Groundwater quality plays a critical role in sustaining agricultural productivity, particularly in semi-arid regions where irrigation heavily depends on groundwater resources. This study presents an integrated framework combining groundwater quality indices (IWQI, EWQI, and GWQI), Geographic Information System (GIS) analysis, and machine learning (ML) models to evaluate the spatio-temporal dynamics of irrigation water quality in the Bahawalpur Irrigation Zone. A dataset comprising 12,026 groundwater samples collected over seven years during pre- and post-monsoon seasons was analyzed using key hydrochemical parameters including EC, SAR, and RSC. Spatial interpolation and seasonal trend analysis revealed a persistent core-rim gradient, where canal-fed central areas exhibited relatively better groundwater quality, while desert-facing peripheral zones showed continuous deterioration due to salinity accumulation and limited recharge. Among the tested ML models, Support Vector Machine (SVM) demonstrated the highest predictive performance, outperforming ANN, Random Forest, and AdaBoost with superior accuracy and generalization ability. The integrated GIS-ML framework effectively identified groundwater quality hotspots and seasonal variability patterns, providing a reliable decision-support tool for sustainable irrigation management and early risk assessment in groundwater-stressed agricultural regions.