Optimizing Urban-Scale Evacuation Strategies Through Disaster Victim Aggregation Modification

As urban areas continue progressing into more complex development, urban areas become more vulnerable to disaster. Therefore, there is growing interest in investigating disaster response processes considering population distribution variation within urban regions. This study explores the unsupervised machine learning methods, K-means, DBSCAN, and introduces a novel Ensemble Clustering, which is compared to the established grid-based method as a framework for aggregating disaster victims in determining evacuation strategies. The Ensemble Clustering method combines the K-Means clustering strengths in dividing disaster victim distribution into globular cluster distribution. DBSCAN supports it as a robust outlier identification for disaggregating the irrelevant member as an individual. Subsequently, the clustering effectivity is evaluated by simulating the evacuation scenarios in the evacuation simulation, and the performance is compared to the traditional grid-based method. Combining the machine learning method and the evacuation simulation can be a framework for the decision-making process during a disaster. The results show that the Ensemble Clustering outperforms the traditional grid method, ensuring a faster completion time.