A Decision Tree Based Hazard Assessment of Karst Sinkholes

Sinkholes are the most common geohazards occurring in east Central Florida (ECF). To understand the sinkhole formation and karst hydrogeology within the ECF region, it is necessary to investigate the sinkhole triggering factors. The main purposes of this study are: (1) to develop a decision-tree based model for sinkhole hazard assessment, which accounts for multiple sinkhole affecting factors; and (2) to construct a sinkhole hazard map of east Central Florida (ECF) using the developed decision tree model. The decision-tree model, a popular classification method in data mining, accounted for five identified factors including geology formation/epoch, overburden thickness, hydraulic head difference, groundwater recharge rate, and proximity to karst features. In this study, the recorded 757 sinkhole locations in ECF were randomly split into 70:30 for training and validation purposes, respectively. The performance test of a produced sinkhole hazard map was carried out with a density analysis where density values should increase from the very low to the very high hazard class.