Application of Nature-Inspired Algorithms to Enhance MLP-Based Water Level Prediction

Accurate forecasting of water level (WL) in coastal rivers is crucial for effective management of water resources and the development of flood hazard mitigation strategies. In Bangladesh, there exists a notable deficiency of insitu datasets pertaining to WLs across multiple rivers. As a result, there is a significant research gap related to the prediction of WLs in coastal rivers in Bangladesh. This study evaluates the efficiency of three nature-inspired algorithms: the Artificial Hummingbird Algorithm (AHA), the Quantum-based Avian Navigation Optimizer Algorithm (QANA), and the Cuckoo Search Algorithm (CS) to improve WL predictions made by the Multilayer Perceptron (MLP) at the Mongla station of Rupsha River. The outcomes of the developed models were checked against four statistical metrics: root mean squared error (RMSE), mean absolute error (MAE), Kling-Gupta Efficiency (KGE), and coefficient of determination ($\mathrm{R}^{2}$). The RMSE and MAE values were found to be lowered significantly by the hybrid CS-MLP model. Additionally, the baseline MLP accuracy was significantly improved by the hybrid AHA-MLP, QANA-MLP, and CS-MLP, respectively. The results of the present study promote efficient modeling of the WL at this station.