Model Optimasi Random Forest dengan PSO-CHI-SM dalam Mengatasi High Dimensional dan Imbalanced Data Banjir Kota Samarinda

Abstract

Flooding is a natural disaster that frequently affects our country. Samarinda City, in particular, continues to experience frequent flooding events with 18 incidents in 2018, 33 incidents in 2020, and 32 incidents in 2021. To predict flood disasters, it is necessary to utilize technology known as machine learning for analyzing and classifying floods. However, classification often encounters issues with high-dimensional data and class imbalance. This study aims to determine the extent to which the accuracy of flood disaster classification improves by using the Random Forest algorithm with PSO for optimization, Chi-Square feature selection, and SMOTE oversampling to balance classes. The data used in this study comprises flood data from 2021-2023 obtained from BMKG and BPBD Samarinda City, with a total of 1095 records and 11 attributes. The validation technique used is 5-fold cross-validation, and the evaluation uses a confusion matrix. The results of the Chi-Square feature selection identified Rainfall, Maximum Wind Direction, Most Frequent Wind Direction, Humidity, Sunshine Duration, and Wind Speed as the most influential features based on Chi-Square scores and P-values. The average accuracy obtained from the proposed classification model using 5-fold cross-validation reached 96.02%.