<p>Developments of artificial intelligence (AI) based models can serve as reliable resources for simulating hydrological processes. Learning algorithms combined with neural network designs are thought to be effective modeling methods for different problems, including groundwater level (GWL) fluctuations. A well-known statistical learning-based method, known as the Relevance Vector Machine (RVM), was combined with an improved metaheuristic algorithm, called improved Manta Ray Foraging Optimization (IMRFO), in this study to create a model that examined the relationship between climatic data (such as precipitation, evapotranspiration, and air temperature), population growth, and GWL changes. The Jammu district’s GWL changes and the primary effective climatic indicators show the strongest association according to the developed model. Using the coefficient of determination (R²), mean absolute error (MAE), index of agreement (IoA), and root mean square error (RMSE), the performance of the chosen models was compared. Findings show that RVM-IMRFO performed better at forecasting than the other models, with a high R² during training data fitting and low RMSE and MAE during testing. The RVM-IMRFO model, which included all significant variables in input scenario 7, yielded the best GWL predictions with R², RMSE, IoA, and MAE values of 0.9849, 0.4961, 0.9881, and 0.4993, respectively, for GWL predictions considering Tanda Sheeda observation well. In addition to providing important insights for developing sustainable groundwater management strategies to guarantee the long-term availability of this essential resource, this study marks a significant milestone in environmental modeling. It concludes that the RVM-IMRFO model produces significantly better predictions of GWL fluctuations.</p>