Sitem Peringatan Dini Banjir Berbasis Machine Learning - Algoritma Random Forest dan Gradient Boosting (Studi Kasus – DAS Ciliwung Jakarta)

Abstract

Flood is one of the natural disasters that frequently occurs in Indonesia, especially during the rainy season. One of the factors that triggers flooding is the overflowing water level of rivers. The use of technology to predict river water levels has been widely implemented. One of the technologies employed is the machine learning technique. This technique can learn patterns from provided data and produce accurate predictions. In this research, a model is designed to predict river water levels using historical data spanning 5 years from the Ciliwung-Cisadane River Basin Agency and BMKG. The dataset undergoes data preprocessing and is then processed using machine learning techniques. The algorithms employed are random forest and gradient boosting algorithms. Both algorithms are assessed in terms of performance by comparing the evaluation metrics RMSE, MAE, and MAPE. The Gradient Boosting algorithm is selected based on its superior performance evaluation, utilizing a parameter combination of n_estimators at 200, max_depth at 5, max_features at 3, min_samples_leaf at 1, and min_samples_split at 4, resulting in MAE and RMSE values of 0.0018 and 0.0163, respectively. With the findings of this research, it is expected to contribute to the development of more accurate river water level prediction technology and aid in making preventive decisions prior to flooding occurrences.