Hybrid Deep Learning Flood Forecasting Framework Optimized with the Snake Algorithm

Abstract

Accurate flood forecasting remains a major challenge due to the nonlinear dynamics of hydrological processes and the difficulty of optimizing deep learning models. This study proposes a hybrid deep learning framework integrating Artificial Neural Networks (ANN), Long Short-Term Memory (LSTM), and Convolutional Neural Networks (CNN) with the Snake Optimization Algorithm (SOA) for hyperparameter tuning. The method includes feature normalization, training–testing partitioning, and multi-metric evaluation using MSE, RMSE, MAE, and R². The results reveal that the hybrid LSTM-SOA model achieved the best performance with R²= 0.8514, MSE=0.000386, RMSE=0.019653, and MAE=0.015849, outperforming standalone models. These results demonstrate the potential of hybrid optimisation-based deep learning as a trustworthy tool to support decisions in flood forecasting, early warning, and disaster preparedness.