Metaheuristic approaches to the placement of suicide bomber detectors

TL;DR

This work addresses OPSBD, the problem of placing δ non-fully reliable detectors on a gridded threat area to minimize expected casualties along potential attacker paths. It introduces a formal attacker-path model with detection probabilities $p_{ijk}=1-\exp(-\eta l_{ijk})$ and casualty function $W$, and compares multiple metaheuristics (HC, GRASP, EA, UMDA) against a greedy baseline. A unified cache and dominance-pruning framework accelerates searches across methods, and extensive experiments on random and real instances establish Hill Climbing as the most consistently effective approach, with clear insights from sensitivity analyses about problem features that drive difficulty. The results have practical implications for rapid and robust deployment of detectors in urban security contexts, and point to promising future directions such as memetic hybrids and longer-time EM-based models for dynamic scenarios.

Abstract

Suicide bombing is an infamous form of terrorism that is becoming increasingly prevalent in the current era of global terror warfare. We consider the case of targeted attacks of this kind, and the use of detectors distributed over the area under threat as a protective countermeasure. Such detectors are non-fully reliable, and must be strategically placed in order to maximize the chances of detecting the attack, hence minimizing the expected number of casualties. To this end, different metaheuristic approaches based on local search and on population-based search are considered and benchmarked against a powerful greedy heuristic from the literature. We conduct an extensive empirical evaluation on synthetic instances featuring very diverse properties. Most metaheuristics outperform the greedy algorithm, and a hill-climber is shown to be superior to remaining approaches. This hill-climber is subsequently subject to a sensitivity analysis to determine which problem features make it stand above the greedy approach, and is finally deployed on a number of problem instances built after realistic scenarios, corroborating the good performance of the heuristic.

Figures (16)

• Figure 1: The bart chart shows the number of casualties (dead and injured) in suicide attacks in the period 2001-2015 (total year figures on top of each bar, scale on the left). The black curve indicates the number of suicide attacks in this time frame (scale on the right). Source: own elaboration based on data from cpost16database.
• Figure 2: An $8 \times 8$ scenario. Green cells correspond to entrances, blue ones to objectives and gray cells represent blocked locations. Lines are shortest paths from each entrance to each objective (note that paths would be extended backwards should a detector be placed on any of the entrances). The dimension of each cell is 10m $\times$ 10m, and the attacker must be neutralized at least 10m away from an objective (we assume that the terrorist moves at a speed of 1m/s and that aborting the attack requires no more than 10 seconds).
• Figure 3: A shortest path going from entrance $e_i$ to objective $o_j$. Areas monitored by detectors $d_k$ and $d_k'$ are those enclosed by circumferences. The segments of this path detected by both detectors are shown with a dotted line and denoted by $l_{ijk}$ and $l_{ijk'}$.
• Figure 4: Dominated cells in the map from Figure \ref{['fig:map3']}, assuming that a total of $\delta=3$ detectors are to be placed. The detector effectiveness radius is $\tau = 10$m. Red bordered cells can be safely disregarded as candidate locations for placing the detectors. Lightest cells are more dominated than darker ones.
• Figure 5: Relative distances to best solutions of results provided by different algorithms on (a) $32 \times 32$ instances and (b) $64 \times 64$ instances.