Predicting where emergencies will happen — and pre-positioning resources to save lives.
In emergency services, minutes determine outcomes. A fire doubles in size every sixty seconds. Cardiac arrest survival drops roughly 10% for each minute without defibrillation. When a crash blocks a highway, secondary incidents — rear-end collisions from slowing traffic — begin within minutes. The fundamental challenge is spatial: emergencies happen across a city, but responders are stationed at fixed locations. If a fire truck is twenty minutes away instead of five, nothing else the system does can make up the difference.
Cities compound this problem by changing constantly. Nashville’s Davidson County has seen unprecedented growth — new subdivisions, commercial corridors, and population shifts that reshape where emergencies occur. Tennessee’s interstate network spans hundreds of miles where a single incident can cascade into hours of gridlock. And urban construction creates its own hazards: Nashville sees roughly 36,000 unauthorized road closures each year, disrupting both traffic and emergency access.
The SCOPE Lab has spent nearly a decade building AI systems that address emergency response across this full arc — from predicting where incidents will happen, through deciding where to station responders, to detecting incidents as they unfold and simulating the operational impact of policy changes. The work has progressed from foundational models through deployed decision-support tools serving fire departments, state DOTs, and city agencies.
Emergency response planning starts with a deceptively simple question: where will the next incident occur? If you can predict spatial and temporal patterns in emergencies — even probabilistically — you can position resources proactively rather than reactively.
Our earliest work approached this through clustering historical incidents and building Bayesian models that identified spatial hotspots and temporal rhythms in 911 call data. Geoff Pettet’s 2017 analysis showed that incident patterns are surprisingly structured: certain intersections, time windows, and weather conditions concentrate risk in ways that statistical models can exploit.
Ayan Mukhopadhyay’s AAMAS 2017 work formalized this as a continuous-time incident prediction problem, moving beyond discrete time bins to model when and where incidents arrive as stochastic processes. This matters because emergency demand does not respect hourly boundaries — a model that predicts “more incidents between 5 and 6 PM” is far less useful than one that estimates arrival rates continuously, enabling dynamic resource repositioning throughout a shift.
Prediction alone is not enough. The harder problem is deciding what to do with a prediction: which unit to send, from where, and how to reposition remaining resources to cover the gaps left behind. This is a sequential decision problem under uncertainty — each dispatch changes the state of the system, and the next incident could arrive anywhere.
The ICCPS 2019 pipeline brought prediction and dispatch together into a unified online decision-theoretic framework. The system continuously updates incident forecasts from streaming data, solves a resource allocation problem to determine where responders should be positioned, and dispatches units to active incidents while accounting for the coverage implications of each assignment. This was the first end-to-end system that connected spatial-temporal prediction directly to operational dispatch decisions.
Geoff Pettet’s AAMAS 2020 work extended this to handle the non-stationarity that plagues real emergency systems — demand patterns shift with seasons, construction, events, and long-term urban change. The approach adapts its decision policies online as the environment evolves, rather than relying on models trained on historical data that may no longer reflect current conditions. This line of research culminated in a hierarchical planning framework for dynamic resource allocation that decomposes the problem across spatial and temporal scales — strategic positioning over hours, tactical repositioning over minutes, and immediate dispatch in seconds.
We also invested in understanding how to design effective decision support systems for emergency operations — because the best algorithm is useless if dispatchers cannot understand, trust, and act on its recommendations.
Individual city models work well, but Tennessee’s emergency infrastructure spans a vast and diverse geography — from Nashville’s dense urban core to hundreds of miles of rural interstate. Scaling incident prediction to this scope required fundamentally different approaches.
Working with the Tennessee Department of Transportation, Sayyed Vazirizade developed large-area incident prediction models that forecast accidents across Tennessee’s interstate network. The challenge is that most road segments see very few incidents — the data is extremely sparse at any individual location. Vazirizade’s approach uses spatial transfer learning to share information across similar road segments, enabling useful predictions even for locations with limited historical data.
This research produced the CRASH Predictive Analytics system, documented in the TDOT project report, which forecasts incident likelihood across the state highway network and optimizes the positioning of TDOT’s HELP (Highway Emergency Lane Patrol) trucks. In evaluation, optimized truck positioning reduced response times by up to 19% and cut average response time by up to 4.5 minutes per incident — significant gains that translate directly into reduced secondary crashes and faster clearance.
The tools developed through this collaboration are publicly available at statresp.ai, which provides the open-source statistical methods and toolkits underlying our emergency response research, and tn.statresp.ai, which provides Tennessee-specific incident analytics and visualization. The StatResp toolkit includes spatial point process models, kernel density estimators, and evaluation frameworks that other researchers and agencies can adapt to their own jurisdictions.
In 2021, we published a comprehensive review in Accident Analysis & Prevention that synthesizes the literature on incident prediction, resource allocation, and dispatch models for emergency management. The survey covers statistical and machine learning approaches to incident forecasting, optimization models for facility location and dynamic repositioning, and dispatch algorithms — providing a unified view of a field that had been fragmented across transportation, operations research, and AI communities. This paper has become a reference point for researchers entering the emergency response domain.
Prediction tells you where incidents are likely; detection tells you where they actually are. The gap between an incident occurring and responders learning about it — the detection delay — is often the largest component of total response time, especially on highways where there may be no witnesses or where 911 calls are imprecise about location.
We explored using crowdsourced data — particularly Waze reports — as a real-time incident detection source. Working with Hemant Purohit at George Mason University, we developed methods for fusing crowdsourced reports with official sensor data to detect incidents faster than either source alone. The ICDM 2021 work advanced this with machine learning models that filter noise from crowdsourced streams and identify genuine incidents with high precision. More recently, we built systems for real-time detection of road closures by integrating Waze data, traffic sensors, and permit records — connecting incident detection directly to the right-of-way monitoring problem.
For Tennessee’s highway network, we developed specialized anomaly detection systems that work directly with traffic sensor data. TRACE uses graph neural networks to model spatial dependencies between highway segments, detecting irregular speed patterns that indicate incidents and localizing where the incident actually is — often miles from where traffic first slows. A complementary approach using Pythagorean mean-based anomaly detection provides lightweight, interpretable incident detection that can run on edge devices with minimal computational resources. Together with TDOT-specific analytics, these systems process data from thousands of sensors across the state road network, providing actionable intelligence to dispatchers and traffic management centers.
Knowing where incidents will happen and detecting them quickly still leaves the stationing problem: where should responders wait between calls? This is a multi-agent coordination challenge — ambulances, fire trucks, and patrol cars must collectively cover a city, and each dispatch creates a coverage gap that other units must compensate for.
Amutheezan Sivagnanam and Ava Pettet developed a multi-agent reinforcement learning approach with hierarchical coordination for emergency responder stationing, presented at ICML 2024. The framework trains agents that learn to reposition cooperatively — not just optimizing their individual coverage, but coordinating to maintain system-wide response capability even as units are dispatched to active incidents and return to service. The hierarchical structure decomposes the problem into district-level coordination and unit-level positioning, making it tractable for city-scale deployment.
Evaluating new stationing and dispatch policies in the real world is risky — a bad policy means slower response times and potentially lives lost. RESPOND (ICCPS 2026) is our incident-level simulation platform for fire and EMS operations that enables safe evaluation of operational policies before field deployment.
RESPOND models the full emergency response cycle: incidents arrive according to learned spatial-temporal distributions, dispatchers assign units according to configurable policies, vehicles travel through a realistic road network with time-varying speeds, and units return to stations after clearing incidents. The simulator couples station placement decisions with dispatch policies — because the best station locations depend on how you dispatch, and vice versa. By running thousands of simulated shifts under different configurations, agencies can compare candidate policies on metrics that matter: average response time, worst-case coverage gaps, unit utilization, and mutual aid frequency.
Nashville’s rapid growth has created a different kind of public safety challenge. The city sees roughly 36,000 unauthorized road closures annually — construction crews blocking lanes without permits, utility work extending beyond approved windows, delivery vehicles occupying travel lanes. These closures cost an estimated $2M in lost permit fees and $5.2M in inspection costs, but more importantly, they create safety risks: blocked lanes force merging, obscure sightlines, and can cut off emergency vehicle access.
We built an urban digital twin — a real-time model of Nashville’s street network integrating LiDAR surveys, street camera feeds, and crowdsourced reports — that detects unauthorized closures automatically and optimizes inspection routes so city crews visit actual violations rather than patrolling randomly. The system was piloted with the Mayor’s Office of Nashville, Nashville DOT, and Metro ITS through a collaboration documented in our access-monitoring work and real-time closure detection research.
This work has now expanded into SENTRY (Sensor-enabled Enforcement of Non-Compliant Traffic Right-of-Way Closures), funded by a $697K NSF CIVIC Innovation Challenge Phase 2 award led by Ayan Mukhopadhyay. SENTRY combines heterogeneous sensor networks — LiDAR, cameras, crowdsourced data — with novel AI/ML algorithms for anomaly detection with limited labeled data, dynamic routing algorithms that blend human and AI decision-making for inspection scheduling, and a real-time scalable software platform for civic infrastructure management. Co-PIs include Daniel Work, Meiyi Ma, and William Barbour at Vanderbilt, working in collaboration with Nashville DOT and Metro Nashville ITS.
Davidson County’s unprecedented growth threatens emergency response performance. Sprawling suburbs, new commercial zones, and population shifts change where emergencies happen — yet fire station locations are fixed, and adding a station is expensive and politically complex.
Working with the Nashville Fire Department and Metro ITS, we developed growth models that project how emergency demand will shift based on population growth, new construction, demographic change, road network expansion, and historical incident patterns. These models estimate future response times under different growth scenarios and identify where new stations would have maximum impact — giving the department evidence-based tools for strategic facility planning rather than relying on intuition or political pressure.
This work reflects collaboration across agencies, institutions, and disciplines. Our partners include the Nashville Fire Department, Tennessee Department of Transportation, the Mayor’s Office of Nashville, Nashville DOT, Metro Nashville ITS, and George Mason University (Hemant Purohit). Key researchers include Ayan Mukhopadhyay, Geoff Pettet, Ava Pettet, Sayyed Vazirizade, Amutheezan Sivagnanam, Ammar Zulqarnain, and William Barbour at Vanderbilt, and Aron Laszka at Penn State.
Funding has been provided by the National Science Foundation (including NSF CIVIC), the Nashville Innovation Alliance, the Tennessee Department of Transportation, and Metro Nashville.
Selected Publications:
@article{mukhopadhyay2021review,
author = {Mukhopadhyay, Ayan and Pettet, Geoffrey and Vazirizade, Sayyed Mohsen and Lu, Di and Jaimes, Alejandro and Said, Said El and Baroud, Hiba and Vorobeychik, Yevgeniy and Kochenderfer, Mykel and Dubey, Abhishek},
journal = {Accident Analysis & Prevention},
title = {A Review of Incident Prediction, Resource Allocation, and Dispatch Models for Emergency Management},
year = {2022},
issn = {0001-4575},
pages = {106501},
volume = {165},
contribution = {lead},
doi = {https://doi.org/10.1016/j.aap.2021.106501},
keywords = {Resource allocation for smart cities, Incident prediction, Computer aided dispatch, Decision making under uncertainty, Accident analysis, Emergency response},
preprint = {https://arxiv.org/abs/2006.04200},
url = {https://www.sciencedirect.com/science/article/pii/S0001457521005327}
}
In the last fifty years, researchers have developed statistical, data-driven, analytical, and algorithmic approaches for designing and improving emergency response management (ERM) systems. The problem has been noted as inherently difficult and constitutes spatio-temporal decision making under uncertainty, which has been addressed in the literature with varying assumptions and approaches. This survey provides a detailed review of these approaches, focusing on the key challenges and issues regarding four sub-processes: (a) incident prediction, (b) incident detection, (c) resource allocation, and (c) computer-aided dispatch for emergency response. We highlight the strengths and weaknesses of prior work in this domain and explore the similarities and differences between different modeling paradigms. We conclude by illustrating open challenges and opportunities for future research in this complex domain.
@inproceedings{iccps2026_respond,
author = {Zulqarnain, Ammar Bin and Talusan, Jose Paolo and Napier, Kelly and Gens, Corey and Higgs, Jennifer and Herndon, Colleen and Mukhopadhyay, Ayan and Dubey, Abhishek},
title = {RESPOND: An Incident-Level Simulation Platform for Fire and EMS Operations},
year = {2026},
booktitle = {Proceedings of the HSCC/ICCPS 2026: 29th ACM International Conference on Hybrid Systems: Computation and Control and 17th ACM/IEEE International Conference on Cyber-Physical Systems},
location = {Saint Malo, France},
keywords = {emergency response, dispatch optimization, facility location, simulation, policy evaluation, urban computing, resource allocation},
note = {Acceptance rate: 28\%; Short Paper; Track: Systems and Applications},
series = {HSCC/ICCPS '26},
what = {RESPOND is a modular cyber-physical system platform for urban emergency response that integrates strategic planning and operational dispatch. The platform provides unified simulation infrastructure for evaluating fire/EMS dispatching policies and station placement strategies under realistic constraints. RESPOND enables scenario-driven evaluation by combining real station operations with hypothetical alternatives, incorporating actual incident data, travel times, and service metrics. The system allows researchers to evaluate counterintuitive policies and trade-offs between competing operational objectives like coverage and response time.},
why = {Urban emergency response is a complex societal-scale CPS involving coordination between multiple agencies, tight operational constraints, and high consequences of failures. Most research remains fragmented and simulation-based, lacking integrated platforms that seamlessly combine strategic planning with operational dispatch evaluation. RESPOND is innovative because it provides a unified testbed for evaluating coupled planning and dispatch decisions at scale, enabling scenario-based exploration of policy alternatives that would be infeasible to test on real systems.},
results = {Simulation fidelity assessment demonstrates that RESPOND accurately reproduces historical incident distributions when using real data, with MAE of approximately 6 incidents per station validating model accuracy. Counterintuitive scenario analysis shows that adding optimally-placed stations near downtown improves coverage by 10-25 seconds while reducing spatial imbalance, demonstrating the platform's ability to reveal non-obvious policy effects.},
project_tags = {emergency, transit, planning, CPS}
}
Growing urban populations strain fire/Emergency Medical Services (EMS) systems, creating societal-scale concerns where decisions about station siting (strategy) and dispatch policies (operations) unfold in a tightly coupled cyber-physical loop. The core challenge lies in validating different approaches since direct experimentation on real populations is infeasible. Prior efforts address isolated components, treating strategic siting heatmaps and operational dispatch heuristics as separate problems. They lack a unified, incident-level simulator to expose the critical cross-policy trade-offs between siting and dispatch. We present RESPOND (REsponse Simulation Platform for Operations, Navigation, and Dispatch), a modular, incident-level, Operational Decision Support System. RESPOND holistically integrates these previously siloed functions, including: (i) optimal station placement, (ii) apparatus allocation, (iii) dispatch policies, (iv) travel time and service time models, and (v) survival modeling for incident prediction. The platform’s engine replays historical incidents at unit resolution and stress-tests counterfactual futures (e.g., station moves, demand surges). A planner-facing interface surfaces key metrics (SLA compliance, 90th Percentile (P90) response time) for deliberation. Evaluations demonstrate reproduction of observed response patterns and reveal policy trade-offs. The result is a unifying platform that transforms fragmented analysis into an operational decision environment, enabling safe and rigorous evaluation of coupled station placement and dispatch policies through simulation.
@inproceedings{10.5555/3692070.3693934,
author = {Sivagnanam, Amutheezan and Pettet, Ava and Lee, Hunter and Mukhopadhyay, Ayan and Dubey, Abhishek and Laszka, Aron},
booktitle = {Proceedings of the 41st International Conference on Machine Learning},
title = {Multi-agent reinforcement learning with hierarchical coordination for emergency responder stationing},
year = {2024},
publisher = {JMLR.org},
series = {ICML'24},
articleno = {1864},
contribution = {colab},
acceptance = {27.5},
location = {Vienna, Austria},
numpages = {22},
url = {https://www.arxiv.org/pdf/2405.13205v1}
}
An emergency responder management (ERM) system dispatches responders, such as ambulances, when it receives requests for medical aid. ERM systems can also proactively reposition responders between predesignated waiting locations to cover any gaps that arise due to the prior dispatch of responders or significant changes in the distribution of anticipated requests. Optimal repositioning is computationally challenging due to the exponential number of ways to allocate responders between locations and the uncertainty in future requests. The state-of-the-art approach in proactive repositioning is a hierarchical approach based on spatial decomposition and online Monte Carlo tree search, which may require minutes of computation for each decision in a domain where seconds can save lives. We address the issue of long decision times by introducing a novel reinforcement learning (RL) approach, based on the same hierarchical decomposition, but replacing online search with learning. To address the computational challenges posed by large, variable-dimensional, and discrete state and action spaces, we propose: (1) actor-critic based agents that incorporate transformers to handle variable-dimensional states and actions, (2) projections to fixed-dimensional observations to handle complex states, and (3) combinatorial techniques to map continuous actions to discrete allocations. We evaluate our approach using realworld data from two U.S. cities, Nashville, TN and Seattle, WA. Our experiments show that compared to the state of the art, our approach reduces computation time per decision by three orders of magnitude, while also slightly reducing average ambulance response time by 5 seconds.
@inproceedings{Mukhopadhyay2019,
author = {Mukhopadhyay, Ayan and Pettet, Geoffrey and Samal, Chinmaya and Dubey, Abhishek and Vorobeychik, Yevgeniy},
booktitle = {Proceedings of the 10th {ACM/IEEE} International Conference on Cyber-Physical Systems, {ICCPS} 2019, Montreal, QC, Canada},
title = {An online decision-theoretic pipeline for responder dispatch},
year = {2019},
pages = {185--196},
bibsource = {dblp computer science bibliography, https://dblp.org},
biburl = {https://dblp.org/rec/bib/conf/iccps/MukhopadhyayPSD19},
category = {selectiveconference},
contribution = {lead},
acceptance = {27},
doi = {10.1145/3302509.3311055},
file = {:Mukhopadhyay2019-An_Online_Decision_Theoretic_Pipeline_for_Responder_Dispatch.pdf:PDF},
keywords = {emergency response, responder dispatch, decision-theoretic planning, SMDP, incident prediction, survival analysis},
project = {smart-cities,smart-emergency-response},
tag = {ai4cps,incident},
timestamp = {Sun, 07 Apr 2019 16:25:36 +0200},
url = {https://doi.org/10.1145/3302509.3311055},
what = {This paper presents an online decision-theoretic pipeline for responder dispatch in emergency management systems. The work formulates the responder dispatch problem as a Semi-Markov Decision Process and develops an online incident prediction model based on survival analysis to enable real-time, data-driven dispatch decisions.},
why = {Emergency response systems face complex challenges in routing limited resources to incidents in dynamic urban environments. Traditional systems dispatch the nearest responder, which ignores future incident probabilities and environmental factors. This paper addresses these limitations through a principled decision-theoretic approach that integrates incident prediction with dynamic dispatch optimization.},
results = {The paper demonstrates the effectiveness of the approach through evaluation on real emergency services data from Nashville, Tennessee. The online prediction and dispatch pipeline reduces response times compared to baseline approaches while accounting for incident cascading effects and changing environmental dynamics. The work successfully bridges incident prediction and optimal dispatch decisions.},
project_tags = {emergency, POMDP, planning, scalable AI}
}
The problem of dispatching emergency responders to service traffic accidents, fire, distress calls and crimes plagues urban areas across the globe. While such problems have been extensively looked at, most approaches are offline. Such methodologies fail to capture the dynamically changing environments under which critical emergency response occurs, and therefore, fail to be implemented in practice. Any holistic approach towards creating a pipeline for effective emergency response must also look at other challenges that it subsumes - predicting when and where incidents happen and understanding the changing environmental dynamics. We describe a system that collectively deals with all these problems in an online manner, meaning that the models get updated with streaming data sources. We highlight why such an approach is crucial to the effectiveness of emergency response, and present an algorithmic framework that can compute promising actions for a given decision-theoretic model for responder dispatch. We argue that carefully crafted heuristic measures can balance the trade-off between computational time and the quality of solutions achieved and highlight why such an approach is more scalable and tractable than traditional approaches. We also present an online mechanism for incident prediction, as well as an approach based on recurrent neural networks for learning and predicting environmental features that affect responder dispatch. We compare our methodology with prior state-of-the-art and existing dispatch strategies in the field, which show that our approach results in a reduction in response time with a drastic reduction in computational time.
@inproceedings{Pettet2020,
author = {Pettet, Geoffrey and Mukhopadhyay, Ayan and Kochenderfer, Mykel and Vorobeychik, Yevgeniy and Dubey, Abhishek},
booktitle = {Proceedings of the 19th Conference on Autonomous Agents and MultiAgent Systems, {AAMAS} 2020, Auckland, New Zealand},
title = {On Algorithmic Decision Procedures in Emergency Response Systems in Smart and Connected Communities},
year = {2020},
category = {selectiveconference},
contribution = {lead},
acceptance = {23},
keywords = {emergency response, decision procedures, resource allocation, algorithmic planning, smart cities, multi-objective optimization},
project = {smart-emergency-response,smart-cities},
tag = {ai4cps, decentralization,incident},
timestamp = {Wed, 17 Jan 2020 07:24:00 +0200},
what = {This paper presents algorithmic decision procedures for emergency response management in smart cities, addressing the problem of optimal incident response under constraints of limited resources and communication disruptions. The work develops both greedy and Monte Carlo Tree Search approaches for dynamically rebalancing emergency responders in response to changing incident patterns. The methodology addresses the tension between minimizing immediate response times and maintaining overall system efficiency.},
why = {Emergency response systems must make decisions about resource allocation in real-time with incomplete information about incident severity and responder availability. Traditional approaches often focus on minimizing response time for individual incidents without considering overall system efficiency or the need to dynamically rebalance resources. This work is innovative because it provides algorithmic approaches for emergency response that optimize over multiple objectives and adapt dynamically to changing incident patterns, enabling more efficient and effective emergency management.},
results = {The algorithmic approaches successfully identify optimal responder allocations and demonstrate that dynamic rebalancing strategies can significantly reduce average response times compared to greedy approaches. Monte Carlo Tree Search provides more sophisticated decision-making by considering future incident probabilities, while greedy approaches offer computational efficiency. Results show that the approach enables emergency response systems to balance immediate response needs with longer-term system efficiency.},
project_tags = {emergency, planning, scalable AI}
}
Emergency Response Management (ERM) is a critical problem faced by communities across the globe. Despite its importance, it is common for ERM systems to follow myopic and straight-forward decision policies in the real world. Principled approaches to aid decision-making under uncertainty have been explored in this context but have failed to be accepted into real systems. We identify a key issue impeding their adoption — algorithmic approaches to emergency response focus on reactive, post-incident dispatching actions, i.e. optimally dispatching a responder after incidents occur. However, the critical nature of emergency response dictates that when an incident occurs, first responders always dispatch the closest available responder to the incident. We argue that the crucial period of planning for ERM systems is not post-incident, but between incidents. However, this is not a trivial planning problem - a major challenge with dynamically balancing the spatial distribution of responders is the complexity of the problem. An orthogonal problem in ERM systems is to plan under limited communication, which is particularly important in disaster scenarios that affect communication networks. We address both the problems by proposing two partially decentralized multi-agent planning algorithms that utilize heuristics and the structure of the dispatch problem. We evaluate our proposed approach using real-world data, and find that in several contexts, dynamic re-balancing the spatial distribution of emergency responders reduces both the average response time as well as its variance.
@article{pettet2021hierarchical,
author = {Pettet, Geoffrey and Mukhopadhyay, Ayan and Kochenderfer, Mykel J. and Dubey, Abhishek},
journal = {ACM Trans. Cyber-Phys. Syst.},
title = {Hierarchical Planning for Dynamic Resource Allocation in Smart and Connected Communities},
year = {2022},
issn = {2378-962X},
month = nov,
number = {4},
volume = {6},
address = {New York, NY, USA},
articleno = {32},
contribution = {lead},
doi = {10.1145/3502869},
issue_date = {October 2022},
keywords = {planning under uncertainty, semi-Markov decision process, large-scale CPS, hierarchical planning, Dynamic resource allocation},
numpages = {26},
preprint = {https://arxiv.org/abs/2107.01292},
publisher = {Association for Computing Machinery},
url = {https://doi-org.proxy.library.vanderbilt.edu/10.1145/3502869}
}
Resource allocation under uncertainty is a classic problem in city-scale cyber-physical systems. Consider emergency response, where urban planners and first responders optimize the location of ambulances to minimize expected response times to incidents such as road accidents. Typically, such problems involve sequential decision making under uncertainty and can be modeled as Markov (or semi-Markov) decision processes. The goal of the decision maker is to learn a mapping from states to actions that can maximize expected rewards. While online, offline, and decentralized approaches have been proposed to tackle such problems, scalability remains a challenge for real world use cases. We present a general approach to hierarchical planning that leverages structure in city level CPS problems for resource allocation. We use emergency response as a case study and show how a large resource allocation problem can be split into smaller problems. We then use Monte Carlo planning for solving the smaller problems and managing the interaction between them. Finally, we use data from Nashville, Tennessee, a major metropolitan area in the United States, to validate our approach. Our experiments show that the proposed approach outperforms state-of-the-art approaches used in the field of emergency response.
@inproceedings{Mukhopadhyay2017,
author = {Mukhopadhyay, Ayan and Vorobeychik, Yevgeniy and Dubey, Abhishek and Biswas, Gautam},
booktitle = {Proceedings of the 16th Conference on Autonomous Agents and MultiAgent Systems, {AAMAS} 2017, S{\~{a}}o Paulo, Brazil, May 8-12, 2017},
title = {Prioritized Allocation of Emergency Responders based on a Continuous-Time Incident Prediction Model},
year = {2017},
acceptance = {27},
pages = {168--177},
bibsource = {dblp computer science bibliography, https://dblp.org},
biburl = {https://dblp.org/rec/bib/conf/atal/MukhopadhyayVDB17},
category = {selectiveconference},
contribution = {colab},
file = {:Mukhopadhyay2017-Prioritized_Allocation_of_Emergency_Responders_based_on_a_Continuous-Time_Incident_Prediction_Model.pdf:PDF},
keywords = {incident prediction, emergency response, optimization, responder allocation, survival analysis},
project = {smart-emergency-response,smart-cities},
tag = {ai4cps,incident},
timestamp = {Wed, 27 Sep 2017 07:24:00 +0200},
url = {http://dl.acm.org/citation.cfm?id=3091154},
what = {This paper addresses emergency responder allocation in urban areas using incident prediction models and optimization algorithms. The work develops methods to predict incident arrival times and severities using survival analysis, then formulates responder allocation as an optimization problem balancing coverage and response times. A hierarchical clustering approach identifies incident patterns.},
why = {Emergency response efficiency depends on optimal responder placement and dispatch. Traditional approaches assume uniform incident distributions and fixed response times. This work is innovative because it combines predictive modeling of incident severity with optimization algorithms to adaptively allocate responders based on spatial-temporal patterns.},
results = {The paper demonstrates improved emergency response times through data-driven responder allocation in Nashville. Results show how incident prediction models enable intelligent dispatch that accounts for incident severity, reducing overall response times compared to traditional approaches.},
project_tags = {emergency, planning, scalable AI, Explainable AI}
}
Efficient emergency response is a major concern in densely populated urban areas. Numerous techniques have been proposed to allocate emergency responders to optimize response times, coverage, and incident prevention. Effective response depends, in turn, on effective prediction of incidents occurring in space and time, a problem which has also received considerable prior attention. We formulate a non-linear mathematical program maximizing expected incident coverage, and propose a novel algorithmic framework for solving this problem. In order to aid the optimization problem, we propose a novel incident prediction mechanism. Prior art in incident prediction does not generally consider incident priorities which are crucial in optimal dispatch, and spatial modeling either considers each discretized area independently, or learns a homogeneous model. We bridge these gaps by learning a joint distribution of both incident arrival time and severity, with spatial heterogeneity captured using a hierarchical clustering approach. Moreover, our decomposition of the joint arrival and severity distributions allows us to independently learn the continuous-time arrival model, and subsequently use a multinomial logistic regression to capture severity, conditional on incident time. We use real traffic accident and response data from the urban area around Nashville, USA, to evaluate the proposed approach, showing that it significantly outperforms prior art as well as the real dispatch method currently in use.
@inproceedings{Pettet2017,
author = {Pettet, Geoffrey and Nannapaneni, Saideep and Stadnick, Benjamin and Dubey, Abhishek and Biswas, Gautam},
booktitle = {2017 {IEEE} SmartWorld},
title = {Incident analysis and prediction using clustering and Bayesian network},
year = {2017},
acceptance = {28},
pages = {1--8},
bibsource = {dblp computer science bibliography, https://dblp.org},
biburl = {https://dblp.org/rec/bib/conf/uic/PettetNSDB17},
category = {selectiveconference},
contribution = {lead},
doi = {10.1109/UIC-ATC.2017.8397587},
file = {:Pettet2017-Incident_analysis_and_prediction_using_clustering_and_Bayesian_network.pdf:PDF},
keywords = {incident prediction, clustering, Bayesian networks, survival analysis, urban analytics},
project = {smart-emergency-response,smart-cities},
tag = {ai4cps,incident},
timestamp = {Wed, 16 Oct 2019 14:14:50 +0200},
url = {https://doi.org/10.1109/UIC-ATC.2017.8397587},
what = {This paper presents a clustering and Bayesian network approach for incident analysis and prediction in urban areas. The work develops unsupervised methods for grouping incidents with similar characteristics and applies survival analysis to predict incident frequencies for specific spatial areas. The methodology integrates data preprocessing, clustering, and probabilistic prediction.},
why = {Incident prediction in large urban areas requires identifying patterns across diverse incident types and locations. Existing approaches often make oversimplified assumptions about incident distributions. This work is innovative because it combines clustering with Bayesian networks to learn incident patterns directly from data.},
results = {The paper demonstrates successful incident prediction for Nashville using real fire department data, achieving significantly higher accuracy than baseline models through cluster-specific prediction. Results show how unsupervised clustering improves prediction accuracy by identifying incident subgroups.},
project_tags = {emergency, ML for CPS, Explainable AI}
}
Advances in data collection and storage infrastructure offer an unprecedented opportunity to integrate both data and emergency resources in a city into a dynamic learning system that can anticipate and rapidly respond to heterogeneous incidents. In this paper, we describe integration methods for spatio-temporal incident forecasting using previously collected vehicular accident data provided to us by the Nashville Fire Department. The literature provides several techniques that focus on analyzing features and predicting accidents for specific situations (specific intersections in a city, or certain segments of a freeway, for example), but these models break down when applied to a large, general area consisting of many road and intersection types and other factors like weather conditions. We use Similarity Based Agglomerative Clustering (SBAC) analysis to categorize incidents to account for these variables. Thereafter, we use survival analysis to learn the likelihood of incidents per cluster. The mapping of the clusters to the spatial locations is achieved using a Bayesian network. The prediction methods we have developed lay the foundation for future work on an optimal emergency vehicle allocation and dispatch system in Nashville.
@inproceedings{vazirizade2021learning,
author = {Vazirizade, Sayyed Mohsen and Mukhopadhyay, Ayan and Pettet, Geoffrey and El Said, Said and Baroud, Hiba and Dubey, Abhishek},
booktitle = {2021 IEEE International Conference on Smart Computing (SMARTCOMP)},
title = {Learning Incident Prediction Models Over Large Geographical Areas for Emergency Response},
year = {2021},
month = aug,
pages = {424-429},
acceptance = {31.7},
contribution = {lead},
doi = {10.1109/SMARTCOMP52413.2021.00091},
issn = {2693-8340},
keywords = {Road accidents;Pipelines;Collaboration;Weather forecasting;Predictive models;Emergency services;Resource management;Spatial Temporal Incident Prediction;Emergency Response Management;Resource Allocation;Statistical Modeling},
tag = {ai4cps,incident}
}
Emergency Response Management (ERM) necessitates the use of models capable of predicting the spatial-temporal likelihood of incident occurrence. These models are used for proactive stationing in order to reduce overall response time. Traditional methods simply aggregate past incidents over space and time; such approaches fail to make useful short-term predictions when the spatial region is large and focused on fine-grained spatial entities like interstate highway networks. This is partially due to the sparsity of incidents with respect to space and time. Further, accidents are affected by several covariates. Collecting, cleaning, and managing multiple streams of data from various sources is challenging for large spatial areas. In this paper, we highlight how this problem is being solved in collaboration with the Tennessee Department of Transportation (TDOT) to improve ERM in the state of Tennessee. Our pipeline, based on a combination of synthetic resampling, clustering, and data mining techniques, can efficiently forecast the spatio-temporal dynamics of accident occurrence, even under sparse conditions. Our pipeline uses data related to roadway geometry, weather, historical accidents, and traffic to aid accident forecasting. To understand how our forecasting model can affect allocation and dispatch, we improve and employ a classical resource allocation approach. Experimental results show that our approach can noticeably reduce response times and the number of unattended incidents in comparison to current approaches followed by first responders. The developed pipeline is efficacious, applicable in practice, and open-source.
@techreport{dot_61069_DS1,
author = {Baroud, Hiba and Dubey, Abhishek and Vazirizade, Sayyed Mohsen and others},
institution = {Tennessee. Department of Transportation},
title = {Collaborative Research Project to Coordinate the Data from the CRASH Predictive Analytics Program Between TDOT and TDOSHS},
year = {2021}
}
@inproceedings{zulqarnain2025,
author = {Zulqarnain, Ammar and Buckelew, Jacob and Talusan, Jose Paolo and Mukhopadhyay, Ayan and Dubey, Abhishek},
booktitle = {2025 IEEE International Conference on Smart Computing (SMARTCOMP)},
title = {TRACE: Traffic Response Anomaly Capture Engine for Localization of Traffic Incidents},
year = {2025},
month = jun,
contribution = {lead},
what = {TRACE is a novel framework for real-time traffic anomaly detection and localization that combines Graph Neural Networks, Transformers, and normalizing flows. The system learns the spatial-temporal dependencies in road networks through graph convolutions while capturing long-range temporal interactions through transformer attention. To detect anomalies, TRACE computes log-likelihoods under a learned probability distribution, identifying points where traffic patterns deviate significantly from normal conditions. The framework provides both anomaly detection and localization through density-based analysis.},
why = {Traditional traffic anomaly detection methods struggle with the complexity of capturing spatial-temporal dependencies in interconnected road networks while maintaining scalability and interpretability. TRACE is innovative because it unifies multiple deep learning paradigms (graph neural networks, transformers, normalizing flows) within a probabilistic framework, enabling unsupervised anomaly detection without requiring labeled anomaly data. The density-based approach provides interpretable anomaly scores grounded in learned probability models.},
results = {Evaluation on real-world traffic data from a mid-sized US metropolitan area demonstrates that TRACE significantly improves incident localization precision by 17% compared to methods that identify anomalies without spatial localization. The framework achieves superior detection latency and mean localization error compared to state-of-the-art baselines.},
keywords = {traffic anomaly detection, graph neural networks, transformers, probabilistic modeling, spatial-temporal analysis, smart transportation, anomaly localization},
project_tags = {CPS, ML for CPS, transit}
}
Effective traffic incident management is critical for road safety and operational efficiency. Yet, many transportation agencies rely on reactionary methods, where incidents are reported by human agents and managed through rule- based frameworks like traditional Traffic Incident Management (TIM) systems. However, these are vulnerable to human error, oversight, and delays during high-stress conditions. Although recent initiatives incorporating real-time sensor data for cor- ridor monitoring and enhanced roadway information systems represent strides toward modernization, these systems often still require substantial human intervention. Recent advancements in graph-based deep learning models offer promising potential for addressing the limitations of traditional methods. While state- of-the-art models exist, the complexities of incident localization within dynamic and interconnected road networks, along with limited availability of high-quality labeled data and variability in real-time traffic measurements, are still open challenges. To address these, we propose the Traffic Response Anomaly Capture Engine (TRACE), a novel approach that combines graph neural networks, transformers, and probabilistic normalizing flows to accurately detect and localize traffic anomalies in real time. TRACE captures spatial-temporal dependencies, manages data uncertainty, and enhances automation, supporting more precise and timely incident localization. Our approach is validated on real-world traffic data and improved incident localization by 0.6 miles (17%) than SOTA methods while maintaining similar incident detection accuracy and mean detection delay.
@article{tcpsislam24,
author = {Islam, Md. Jaminur and Talusan, Jose Paolo and Bhattacharjee, Shameek and Tiausas, Francis and Dubey, Abhishek and Yasumoto, Keiichi and Das, Sajal K.},
journal = {ACM Trans. Cyber-Phys. Syst.},
title = {Scalable Pythagorean Mean-based Incident Detection in Smart Transportation Systems},
year = {2024},
issn = {2378-962X},
month = may,
number = {2},
volume = {8},
address = {New York, NY, USA},
articleno = {20},
contribution = {colab},
doi = {10.1145/3603381},
issue_date = {April 2024},
keywords = {Weakly unsupervised learning, anomaly detection, smart transportation, graph algorithms, cluster analysis, regression, incident detection, approximation algorithm},
numpages = {25},
publisher = {Association for Computing Machinery},
url = {https://doi.org/10.1145/3603381}
}
Modern smart cities need smart transportation solutions to quickly detect various traffic emergencies and incidents in the city to avoid cascading traffic disruptions. To materialize this, roadside units and ambient transportation sensors are being deployed to collect speed data that enables the monitoring of traffic conditions on each road segment. In this article, we first propose a scalable data-driven anomaly-based traffic incident detection framework for a city-scale smart transportation system. Specifically, we propose an incremental region growing approximation algorithm for optimal Spatio-temporal clustering of road segments and their data; such that road segments are strategically divided into highly correlated clusters. The highly correlated clusters enable identifying a Pythagorean Mean-based invariant as an anomaly detection metric that is highly stable under no incidents but shows a deviation in the presence of incidents. We learn the bounds of the invariants in a robust manner such that anomaly detection can generalize to unseen events, even when learning from real noisy data. Second, using cluster-level detection, we propose a folded Gaussian classifier to pinpoint the particular segment in a cluster where the incident happened in an automated manner. We perform extensive experimental validation using mobility data collected from four cities in Tennessee and compare with the state-of-the-art ML methods to prove that our method can detect incidents within each cluster in real-time and outperforms known ML methods.
@techreport{barbour2024tdot,
author = {Barbour, William and Baroud, Hiba and Dubey, Abhishek and Sprinkle, Jonathan and Work, Daniel},
title = {TDOT RDS Data Quality Assurance and High-Resolution Content Enhancement},
year = {2024},
url = {https://trid.trb.org/View/2499199}
}
@article{10.1145/3633784,
author = {Senarath, Yasas and Mukhopadhyay, Ayan and Purohit, Hemant and Dubey, Abhishek},
journal = {Digit. Gov.: Res. Pract.},
title = {Designing a Human-centered AI Tool for Proactive Incident Detection Using Crowdsourced Data Sources to Support Emergency Response},
year = {2024},
month = mar,
number = {1},
volume = {5},
address = {New York, NY, USA},
articleno = {9},
contribution = {colab},
doi = {10.1145/3633784},
issue_date = {March 2024},
keywords = {Emergency response, incident detection, human-centered ai tool, crowdsourcing},
numpages = {19},
publisher = {Association for Computing Machinery},
url = {https://doi.org/10.1145/3633784}
}
Time of incident reporting is a critical aspect of emergency response. However, the conventional approaches to receiving incident reports have time delays. Non-traditional sources such as crowdsourced data present an opportunity to detect incidents proactively. However, detecting incidents from such data streams is challenging due to inherent noise and data uncertainty. Naively maximizing detection accuracy can compromise spatial-temporal localization of inferred incidents, hindering response efforts. This article presents a novel human-centered AI tool to address the above challenges. We demonstrate how crowdsourced data can aid incident detection while acknowledging associated challenges. We use an existing CROME framework to facilitate training and selection of best incident detection models, based on parameters suited for deployment. The human-centered AI tool provides a visual interface for exploring various measures to analyze the models for the practitioner’s needs, which could help the practitioners select the best model for their situation. Moreover, in this study, we illustrate the tool usage by comparing different models for incident detection. The experiments demonstrate that the CNN-based incident detection method can detect incidents significantly better than various alternative modeling approaches. In summary, this research demonstrates a promising application of human-centered AI tools for incident detection to support emergency response agencies.
@inproceedings{ICDM_2021,
author = {Senarath, Yasas and Mukhopadhyay, Ayan and Vazirizade, Sayyed and hemant Purohit and Nannapaneni, Saideep and Dubey, Abhishek},
booktitle = {21st IEEE International Conference on Data Mining (ICDM 2021)},
title = {Practitioner-Centric Approach for Early Incident Detection Using Crowdsourced Data for Emergency Services},
year = {2021},
acceptance = {20},
contribution = {colab},
tag = {ai4cps,incident},
what = {This paper presents CROME, a crowdsourced multi-objective event detection framework for early incident detection using crowdsourced data from Waze and traffic incident reports. The system balances the conflicting objectives of spatial-temporal accuracy and temporal responsiveness for incident detection. The approach uses convolutional neural networks and multi-objective optimization to find Pareto-optimal solutions.},
why = {Early incident detection from crowdsourced data is important for emergency response but must balance the competing objectives of detection accuracy and responsiveness. This work is significant because it formulates incident detection as a multi-objective optimization problem that enables explicit trade-off analysis. The approach demonstrates how to leverage noisy crowdsourced data while maintaining principled reasoning about accuracy-responsiveness trade-offs.},
results = {The CROME framework was evaluated on real traffic incident data from Nashville and demonstrated superior performance compared to single-objective baseline approaches. The multi-objective optimization identified Pareto-optimal solutions that practitioners can select based on their priorities. The system successfully detected incidents significantly earlier than traditional methods while maintaining acceptable spatial accuracy.},
keywords = {incident detection, crowdsourced data, multi-objective optimization, emergency response, traffic monitoring},
project_tags = {emergency, transit, ML for CPS, scalable AI}
}
Emergency response is highly dependent on the time of incident reporting. Unfortunately, the traditional approach to receiving incident reports (e.g., calling 911 in the USA) has time delays. Crowdsourcing platforms such as Waze provide an opportunity for early identification of incidents. However, detecting incidents from crowdsourced data streams is difficult due to the challenges of noise and uncertainty associated with such data. Further, simply optimizing over detection accuracy can compromise spatial-temporal localization of the inference, thereby making such approaches infeasible for real-world deployment. This paper presents a novel problem formulation and solution approach for practitioner-centered incident detection using crowdsourced data by using emergency response management as a case-study. The proposed approach CROME (Crowdsourced Multi-objective Event Detection) quantifies the relationship between the performance metrics of incident classification (e.g., F1 score) and the requirements of model practitioners (e.g., 1 km. radius for incident detection). First, we show how crowdsourced reports, ground-truth historical data, and other relevant determinants such as traffic and weather can be used together in a Convolutional Neural Network (CNN) architecture for early detection of emergency incidents. Then, we use a Pareto optimization-based approach to optimize the output of the CNN in tandem with practitioner-centric parameters to balance detection accuracy and spatial-temporal localization. Finally, we demonstrate the applicability of this approach using crowdsourced data from Waze and traffic accident reports from Nashville, TN, USA. Our experiments demonstrate that the proposed approach outperforms existing approaches in incident detection while simultaneously optimizing the needs for realworld deployment and usability.
@inproceedings{senarath_emergency_2020,
author = {Senarath, Yasas and Nannapaneni, Saideep and Purohit, Hemant and Dubey, Abhishek},
booktitle = {The 2020 {IEEE}/{WIC}/{ACM} {International} {Joint} {Conference} {On} {Web} {Intelligence} {And} {Intelligent} {Agent} {Technology}},
title = {Emergency {Incident} {Detection} from {Crowdsourced} {Waze} {Data} using {Bayesian} {Information} {Fusion}},
year = {2020},
month = nov,
note = {arXiv: 2011.05440},
publisher = {IEEE},
acceptance = {30},
annote = {Comment: 8 pages, The 2020 IEEE/WIC/ACM International Joint Conference On Web Intelligence And Intelligent Agent Technology (WI-IAT '20)},
contribution = {colab},
copyright = {All rights reserved},
file = {arXiv Fulltext PDF:/Users/abhishek/Zotero/storage/B8WHQRUX/Senarath et al. - 2020 - Emergency Incident Detection from Crowdsourced Waz.pdf:application/pdf;arXiv.org Snapshot:/Users/abhishek/Zotero/storage/98PX572Y/2011.html:text/html},
keywords = {Computer Science - Artificial Intelligence, Computer Science - Social and Information Networks},
tag = {incident},
url = {http://arxiv.org/abs/2011.05440},
urldate = {2021-01-31}
}
The number of emergencies have increased over the years with the growth in urbanization. This pattern has overwhelmed the emergency services with limited resources and demands the optimization of response processes. It is partly due to traditional ‘reactive’ approach of emergency services to collect data about incidents, where a source initiates a call to the emergency number (e.g., 911 in U.S.), delaying and limiting the potentially optimal response. Crowdsourcing platforms such as Waze provides an opportunity to develop a rapid, ‘proactive’ approach to collect data about incidents through crowd-generated observational reports. However, the reliability of reporting sources and spatio-temporal uncertainty of the reported incidents challenge the design of such a proactive approach. Thus, this paper presents a novel method for emergency incident detection using noisy crowdsourced Waze data. We propose a principled computational framework based on Bayesian theory to model the uncertainty in the reliability of crowd-generated reports and their integration across space and time to detect incidents. Extensive experiments using data collected from Waze and the official reported incidents in Nashville, Tenessee in the U.S. show our method can outperform strong baselines for both F1-score and AUC. The application of this work provides an extensible framework to incorporate different noisy data sources for proactive incident detection to improve and optimize emergency response operations in our communities.
@inproceedings{pettet2022designing,
author = {Pettet, G. and Baxter, H. and Vazirizade, S. and Purohit, H. and Ma, M. and Mukhopadhyay, A. and Dubey, A.},
booktitle = {2022 Workshop on Cyber Physical Systems for Emergency Response (CPS-ER)},
title = {Designing Decision Support Systems for Emergency Response: Challenges and Opportunities},
year = {2022},
address = {Los Alamitos, CA, USA},
month = may,
pages = {30-35},
publisher = {IEEE Computer Society},
contribution = {lead},
doi = {10.1109/CPS-ER56134.2022.00012},
keywords = {emergency response, resource allocation, incident detection, incident forecasting, cyber-physical systems},
url = {https://doi.ieeecomputersociety.org/10.1109/CPS-ER56134.2022.00012},
what = {This paper presents a comprehensive framework for designing emergency response management systems, addressing the challenges of coordinating multiple agencies, collecting diverse geospatial data, and providing incident forecasting and resource allocation. The system integrates data curation components, incident detection models, and dynamic resource allocation algorithms while emphasizing how to handle sparse and uncertain incident data across large geographic areas.},
why = {Emergency response systems must operate under conditions of uncertainty and incomplete information while coordinating many heterogeneous agencies and data sources. This work is innovative because it provides a principled approach that integrates multiple technical challenges including data integration, learning under sparsity, and decentralized decision-making. The authors address the fundamental tension between needing accurate predictions and having limited incident data.},
results = {The framework was evaluated on real incident data from Tennessee highways and demonstrated superior performance compared to existing approaches. Key findings show that incident prediction models benefit significantly from geographic and temporal feature engineering, and that resource allocation must adapt to non-stationary incident patterns. The system successfully integrated sparse incident data with weather and traffic information to improve emergency response times.},
project_tags = {emergency, planning, CPS, scalable AI}
}
Designing effective emergency response management (ERM) systems to respond to incidents such as road accidents is a major problem faced by communities. In addition to responding to frequent incidents each day (about 240 million emergency medical services calls and over 5 million road accidents in the US each year), these systems also support response during natural hazards. Recently, there has been a consistent interest in building decision support and optimization tools that can help emergency responders provide more efficient and effective response. This includes a number of principled subsystems that implement early incident detection, incident likelihood forecasting and strategic resource allocation and dispatch policies. In this paper, we highlight the key challenges and provide an overview of the approach developed by our team in collaboration with our community partners.
@article{wilburaccess2020,
author = {{Talusan}, J. P. V. and {Wilbur}, M. and {Dubey}, A. and {Yasumoto}, K.},
journal = {IEEE Access},
title = {Route Planning Through Distributed Computing by Road Side Units},
year = {2020},
pages = {176134-176148},
volume = {8},
contribution = {minor},
tag = {decentralization,transit}
}
Cities are embracing data-intensive applications to maximize their constrained transportation networks. Platforms such as Google offer route planning services to mitigate the effect of traffic congestion. These use remote servers that require an Internet connection, which exposes data to increased risk of network failures and latency issues. Edge computing, an alternative to centralized architectures, offers computational power at the edge that could be used for similar services. Road side units (RSU), Internet of Things (IoT) devices within a city, offer an opportunity to offload computation to the edge. To provide an environment for processing on RSUs, we introduce RSU-Edge, a distributed edge computing system for RSUs. We design and develop a decentralized route planning service over RSU-Edge. In the service, the city is divided into grids and assigned an RSU. Users send trip queries to the service and obtain routes. For maximum accuracy, tasks must be allocated to optimal RSUs. However, this overloads RSUs, increasing delay. To reduce delays, tasks may be reallocated from overloaded RSUs to its neighbors. The distance between the optimal and actual allocation causes accuracy loss due to stale data. The problem is identifying the most efficient allocation of tasks such that response constraints are met while maintaining acceptable accuracy. We created the system and present an analysis of a case study in Nashville, Tennessee that shows the effect of our algorithm on route accuracy and query response, given varying neighbor levels. We find that our system can respond to 1000 queries up to 57.17% faster, with only a model accuracy loss of 5.57% to 7.25% compared to using only optimal grid allocation.
@inproceedings{Pettet2019,
author = {Pettet, Geoffrey and Mukhopadhyay, Ayan and Samal, Chinmaya and Dubey, Abhishek and Vorobeychik, Yevgeniy},
booktitle = {Proceedings of the 10th {ACM/IEEE} International Conference on Cyber-Physical Systems, {ICCPS} 2019, Montreal, QC, Canada},
title = {Incident management and analysis dashboard for fire departments: {ICCPS} demo},
year = {2019},
pages = {336--337},
bibsource = {dblp computer science bibliography, https://dblp.org},
biburl = {https://dblp.org/rec/bib/conf/iccps/PettetMSDV19},
category = {poster},
contribution = {lead},
doi = {10.1145/3302509.3313329},
file = {:Pettet2019-Incident_management_and_analysis_dashboard_for_fire_departments_ICCPS_demo.pdf:PDF},
keywords = {emergency management, incident analysis, prediction, dashboard, visualization, spatial-temporal data},
project = {smart-cities,smart-emergency-response},
tag = {incident},
timestamp = {Sun, 07 Apr 2019 16:25:36 +0200},
url = {https://doi.org/10.1145/3302509.3313329},
what = {This paper presents a dashboard tool for analyzing and managing spatial-temporal incidents in emergency response systems. The work integrates incident prediction models with interactive visualization capabilities to help emergency managers analyze historical incident distributions and plan resource deployment.},
why = {Emergency response systems require decision support tools that integrate data analytics with situational awareness. Existing tools often separate historical analysis from predictive modeling and dispatch planning. This dashboard integrates survival analysis-based incident prediction with interactive maps and statistical visualizations to enable comprehensive incident analysis and planning.},
results = {The paper demonstrates the dashboard through a case study analyzing incidents from Nashville's emergency services. The system displays historical incident density, predicted future incident distributions, and enables exploration of depot effects on response times. The interactive tool successfully integrates incident prediction with spatial planning capabilities.},
project_tags = {emergency, planning, scalable AI}
}
This work presents a dashboard tool that helps emergency responders analyze and manage spatial-temporal incidents like crime and traffic accidents. It uses state-of-the-art statistical models to learn incident probabilities based on factors such as prior incidents, time and weather. The dashboard can then present historic and predicted incident distributions. It also allows responders to analyze how moving or adding depots (stations for emergency responders) affects average response times, and can make dispatching recommendations based on heuristics. Broadly, it is a one-stop tool that helps responders visualize historical data as well as plan for and respond to incidents.