Workplace safety is under serious threat due to the increasing trend of active shooter violence in recent years. Therefore, it becomes essential that the safety of a workplace is rigorously and, most importantly, methodologically assessed against active shooter violence. To serve this purpose, this study proposes a machine learning-optimization framework to assess the safety of a building against possible active shooter violence. First, several state-of-the-art machine learning models are employed to predict an agent’s movement decisions (with directions) under different violence scenarios. The predictions are then utilized in a mixed-integer linear programming model to maximize the agent’s utility under a possible active shooter violence situation. The machine learning models and the proposed optimization model considered several building-specific (e.g., staircase/hiding room capacities, building orientation) and agent-specific (e.g., herding behavior, cognitive delay) attributes to realistically capture the violent situation. The performance of the proposed machine learning-optimization framework is assessed on a two-storied test building. Results indicate that the building configuration (e.g., number and location of the staircases, hiding rooms, exits) as well as agent behaviors, such as herding behavior and cognitive delay, play an important role in the recovery/casualty of civilians under a crisis situation.

This study proposes several decision-making tools utilizing optimization and machine learning frameworks to assess and improve the safety of the workplaces. The first chapter of this study presents a novel mathematical model to optimally locate a set of detectors to minimize the expected number of casualties in a given threat area. The problem is formulated as a nonlinear binary integer programming model and then solved as a linearized branch-and-bound algorithm. Several sensitivity analyses illustrate the model's robustness and draw key managerial insights. One of the prevailing threats in the last decades, Active Shooting (AS) violence, poses a serious threat to public safety. The second chapter proposes an innovative mathematical model which captures several essential features (e.g., the capacity of the facility and individual choices, heterogeneity of individual behavioral and choice sets, restriction on choice sets depending on the location of the shooter and facility orientation, and many others) which are essential for appropriately characterizing and analyzing the response strategy for civilians under an AS exposed environment. We demonstrate the applicability of the proposed model by implementing the effectiveness of the RUN.HIDE.FIGHT.® (RHF) program in an academic environment. Given most of the past incidents took place in built environments (e.g., educational and commercial buildings), there is an urgent need to methodologically assess the safety of the buildings under an active shooter situation. Finally, the third chapter aims to bridge this knowledge gap by developing a learning technique that can be used to model the behavior of the shooter and the trapped civilians in an active shooter incident. Understanding how the civilians responded to different simulated environments, a number of actions could have been undertaken to bolster the safety measures of a given facility. Finally, this study provides a customized decision-making tool that adopts a tailored maximum entropy inverse reinforcement learning algorithm and utilizes safety measurement metrics, such as the percentage of civilians who can hide/exit in/from the system, to assess a workplace's safety under an active shooter incident.

This dissertation investigates the value of deep reinforcement learning (DRL) within an agent-based model (ABM) of a large open-air venue. The intent is to reduce civilian casualties in an active shooting incident (ASI). There has been a steady increase of ASIs in the United States of America for over 20 years, and some of the most casualty-producing events have been in open spaces and open-air venues. More research should be conducted within the field to help discover policies that can mitigate the threat of a shooter in extremis. This study uses the concept of dynamic signage, controlled by a DRL policy, to guide civilians away from the threat and toward a safe exit in the modeled environment. It was found that a well-trained DRL policy can significantly reduce civilian casualties as compared to baseline scenarios. Further, the DRL policy can assist decision makers in determining how many signs to use in an environment and where to place them. Finally, research using DRL in the ASI space can yield systems and policies that will help reduce the impact of active shooters during an incident.