<![CDATA[Floods are hydrometeorological disasters that often occur in tropical regions such as Indonesia and can have significant impacts on infrastructure, economy, and public health. This study aims to build and compare the performance of 21 artificial intelligence models, consisting of 15 Machine Learning algorithms and 6 Deep Learning architectures, in classifying flood risk levels based on multivariate tabular data. The dataset used includes 22 relevant environmental and social variables, with classification targets in four classes: Low, Moderate, High, and Very High. To improve data quality, feature selection was carried out using the LASSO method and class balancing with the SMOTEENN technique. The evaluation results showed that the C4.5, MLP, Random Forest, and Logistic Regression models obtained the highest accuracy (>94%), followed by deep learning models such as BiLSTM, CNN, and BiGRU with competitive accuracy (≥90%). Confusion matrix analysis confirmed the consistency of predictions across classes with a balanced distribution, especially in the decision tree and deep neural network models. This study emphasizes the importance of selecting a model that suits the characteristics of the data to achieve optimal predictions. The pipeline developed in this study is expected to be the basis for a more accurate and adaptive AI-based early warning system in mitigating flood risks in the future.]]>
Copyrights © 2025
