Employing Machine Learning for Flood Prediction: A Comparative Study of Tree-Based Techniques

Flooding is one of the most serious problems in urban areas, especially in cities like Bangkok, and is further exacerbated by rapid urbanization and geographic conditions. This study proposes a technique to predict flood occurrence and forecast flood level for Bangkok using tree-based machine learning models, including Random Forest, Light Gradient Boosting Machine (LightGBM), and eXtreme Gradient Boosting (XGBoost). The paper presents a developed comprehensive dataset, integrating meteorological features and temporal features for model training and testing. The experimental results show that XGBoost attained the highest accuracy of flood occurrence prediction, <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 1. 4 2 \%}$</tex>, while the best performance in the prediction of flood level was with Random Forest, presenting the lowest error metrics: MSE of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$39.33 \mathbf{~ c m}^{2}$</tex> and MAE of 4.43 cm. The feature importance analysis revealed that rainfall, relative humidity, maximum temperature, and wind speed were the most influential features, accounting for about 74.89 % of flood occurrence prediction. The proposed model provides useful insights into urban planning and disaster management, besides underlining the potential of machine learning for the accurate prediction of floods in highly urbanized regions.