The study develops an integrated geospatial data-driven framework to map groundwater potential in semi-arid Agra, addressing urbanization-induced stress on aquifers and the limitations of ground-based assessments for regional planning. Using multicollinearity-screened factors-rainfall, elevation, slope, drainage, geology, LULC, and soil-the workflow combines AHP-based multi-criteria weighting with machine learning techniques, converting inputs into raster layers and validating outputs with groundwater well locations and water-level data. Rainfall emerged as the dominant predictor, contributing up to 75.17% in Random Forest and 75% in Boosted Regression Tress feature importance, underscoring monsoonal control on recharge. Model performance assessed via ROC-AUC showed RF (81.6% pre-monsoon, 79.9% post-monsoon) and BRT (81.4%, 80.7%) substantially outperform AHP (70.1%, 70.0%), indicating stronger discrimination from data-driven learning. Spatially, high groundwater potential concentrates in the middle-eastern Agra corridor between the Yamuna and Chambal rivers, reaching 32.93% areal extent in post-monsoon RF results, while southwestern and western tracts consistently remain very low to low across all models. Seasonal groundwater-level analysis reveals post-monsoon depth reduction-most pronounced in recharge-favorable zones-with observed ranges of 4.78-50.7 m bgl (pre-monsoon) and 3.8-50.5 m bgl (post-monsoon). The hybrid MCDA-ML framework offers a robust, scalable basis for groundwater zoning, guiding sustainable extraction, recharge targeting, and urban-peri-urban water management.