Groundwater Resource Prediction and Management Using Comment Feedback Optimization Algorithm for Deep Learning

Water resource management remains a critical challenge globally, exacerbated by climate change, urbanization, and inefficient resource allocation. Predicting groundwater availability plays a crucial role in mitigating these challenges, especially in regions facing water scarcity. This study leverages the Comment Feedback Optimization Algorithm (CFOA) to enhance deep learning models for groundwater resource prediction. We propose a binary version of CFOA (bCFOA) for feature selection, which significantly improves the prediction performance of the LightGBM (Light Gradient Boosting Machine) model. The baseline LightGBM model achieved a mean squared error (MSE) of 0.045470041. After applying bCFOA for feature selection, the model’s performance improved with an average error of 0.41055. Furthermore, the integration of CFOA for hyperparameter optimization of LightGBM resulted in an impressive MSE of 6.11673E-06, demonstrating substantial improvements in accuracy. To further assess the robustness, efficiency, and interpretability of the proposed framework, an extensive ablation study was conducted. This included sensitivity analysis of key hyperparameters ( Z 0 , K 0 , λ), complexity evaluation of various metaheuristic variants, and cluster-based visual diagnostics. The study revealed not only the importance of parameter stability and search behavior, but also offered evidence of CFOA’s superior convergence and resource efficiency relative to competing approaches. These results highlight the potential of CFOA and bCFOA in optimizing machine learning models for groundwater resource prediction, offering valuable insights for sustainable water management. The added ablation insights strengthen the generalizability of the proposed method, illustrating its adaptability to broader environmental modeling scenarios. The implications of this work are far-reaching, suggesting that CFOA-based optimization can be applied to various environmental modeling tasks, enhancing predictive capabilities and facilitating more efficient resource management practices.