Optimasi Model Algoritma Machine Learning Suppervised menggunakan Algoritma Genetika untuk Prediksi Kebakaran Hutan dan Lahan

Abstract

Forest and land fires are a common occurrence in Indonesia, particularly in the provinces of Sumatra and Kalimantan. One strategy for mitigating the impact of forest and land fires is to predict areas that are prone to such incidents. In this study, genetic algorithm (GA) optimization was employed to enhance the efficacy of the random tree and hyper-SVM algorithms, with a view to identifying the most optimal test results. The dataset utilized in this study comprises hotspot data and climate data for Ketapang Regency during the 2021-2022 period. The results of the training and testing demonstrate that the Random Tree +GA algorithm optimization with a PC value of 0.6 and Bolzmann selection parameters yields an accuracy of 99.77%, a recall of 94.88%, a precision of 95%, an RMSE of 0.015, and a Kappa of 0.9. In contrast, the Hyper-SVM +GA optimization, with a PC value of 0.6 and Bolzmann selection parameters, yielded an accuracy of 70.48%, a recall of 52.14%, a precision of 50.58%, an RMSE of 0.493, and a Kappa of 0.026. The results demonstrate that the Random Tree +GA algorithm model optimization exhibits superior performance compared to Hyper-SVM +GA optimization. Consequently, it can be inferred that the Random Tree +GA algorithm represents the most effective classification model for predicting the likelihood of forest and land fires in Ketapang Regency