Victim clustering with k-prototype algorithm for flood evacuation planning

Global warming intensifies inevitable severe floods, thus necessitating robust evacuation planning to minimize disaster impacts through swift assistance. Recognizing the interconnectedness of demand and supply aspects, effective evacuation planning considers evacuee behavior through victim clustering, which is of utmost importance. Despite previous efforts in modeling victim behavior, there remains a gap in incorporating victim clustering explicitly in flood evacuation planning. Thus, this study aims to adopt k-prototype algorithm, which is capable of handling mixed-type features, to perform victim clustering for probable flood occurrence by considering numerous influential factors including risk perception, compliance level and arrival pattern of victims. The k-prototype clustering was performed via McClain index (for performance assessment) on an illustrative mixed-type dataset (with 10 variables) comprising 498 valid respondents, for the context of Central Region of Malaysia. The findings show that the optimal number of clusters, which ranges from 2 to 5, could be formed effectively for 8 distinct scenarios. Besides, it is noticeable that the probable response to evacuate (somewhat and very likely) is about 52% under uncertainty. Concisely, this study aspires to furnish emergency planners with beneficial insights in implementing effective evacuation strategies to reduce the negative impact of flood occurrence significantly.