Samuel K. Afful, Cyril Dziedzorm Boateng, Emmanuel Ahene, Jeffery Nii Armah Aryee, David D. Wemegah, Solomon S.R. Gidigasu, Akyana Britwum, Marian A. Osei, Jesse Gilbert , Haoulata Touré, Vera Mensah

Physical models have long been employed for groundwater level (GWL) prediction. Recently, artificial intelligence (AI), particularly neural networks (NNs), have gained widespread use in forecasting GWL. Forecasting of GWL is essential to enable analyze, quantify and manage groundwater. This systematic review investigates the application of NNs for GWL prediction, focusing on the architectures of the various NN models employed. The study utilizes the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) methodology to screen and synthesize relevant scientific article. Various NN architectures, such as artificial neural networks (ANNs), feedforward neural networks (FFNNs), backpropagation neural networks (BPNNs), long short-term memory (LSTM), and hybrid models, were analyzed. The results from the systematic review indicates a growing preference for hybrid models, which are effective in capturing hidden relationships between GWL and environmental factors. The root mean square error (RMSE) emerges as the predominant performance metrics, highlighting its significance in evaluating NNs. Results from the review also highlights the significance of comprehensive, long-term datasets covering a decade for a robust trend analyses and accurate predictions. The findings, contribute to a deeper understanding of new trends in groundwater research such as the application of neural networks for prediction problems in groundwater research. In conclusion, hybrid metaheuristic algorithm produced more efficient results emphasizing their efficacy. In addition, lagged values were essential input for GWL prediction. The paper, addressed both technical nuances and broader environmental implications