Classification of Flood-Prone Areas Using 10-Fold Cross Validation and K-Nearest Neighbors

Abstract

Floods occur when the volume of water exceeds the capacity of containment or the normal flow of water channels, such as rivers, lakes, canals, or drainage systems. In 2021 and 2022, there were 88 flood disasters in Semarang. Factors influencing the occurrence of flood disasters include rainfall, land slope, and the vegetation cover index (NDVI). The main objective of this research is to identify and classify areas in Semarang that are susceptible to flooding by mapping them based on rainfall, vegetation index, and land slope. The classified statuses are flood-prone areas and non-flood-prone areas. A spatial assessment of areas prone to flooding based on GIS is performed using the K-Nearest Neighbor (K-NN) classification algorithm with the R programming language to accomplish this objective. Optimal k testing using the 10-fold cross-validation method reveals that the highest accuracy of the system is achieved with the K-NN method at k=7, reaching 86%. The findings from this study suggest that eight sub-district areas within Semarang are susceptible to flooding, whereas 167 areas in Semarang do not exhibit susceptibility to floods. The potential areas prone to flooding are represented by making a Semarang flood susceptibility map using Quantum GIS software.