Groundwater from coastal aquifers plays a significant role in agriculture, but its diminishing of quality often impacts crop production and soil sustainability by leading to soil salinization and the deterioration of irrigation water standards. This study addresses the pressing issue at Mornang Plain in Tunisia utilizing an integrated approach that combines statistical analysis (principal component analysis (PCA) and cluster analysis (CA), geographic information system (GIS), and machine learning (ML) techniques to assess and predict irrigation water quality. Key parameters such as irrigation water quality index (IWQI), potential salinity (PS), sodium percentage (Na%), and sodium adsorption ratio (SAR) were evaluated to assess water quality for agricultural use. The study identified three main groundwater facies (Na-Cl, Ca-Mg-SO4, Ca-Mg-Cl/SO4), that displaying distinct chemical signatures shaped by geological, hydrological, and human processes. The analysis showed that over 65% of the groundwater samples fall within the "unsuitable" category for irrigation, with high to severe constraints for soil and crop sustainability. A novel decision tree (DT) based ML model was optimized to predict these irrigation indices, achieving high performance with fewer input parameters. With low RMSE values and R2 values ranging from 0.706 to 0.996 across several indices, the DT models showed remarkable predictive accuracy. The models' efficiency in producing accurate water quality forecasts at lower analytical costs is demonstrated by their R2 = 0.992 (RMSE = 1.693) for IWQI and 0.996 (RMSE = 0.822) for PS. This approach provides a cost-effective alternative to traditional methods by reducing the number of chemical parameters required for analysis. The results of this study offer significant insights for water resource management in arid and semi-arid regions, highlighting the potential of ML techniques in predicting irrigation water quality. The findings are valuable not only for Tunisia but also for similar regions worldwide, offering a tool for decision-makers to develop sustainable water management strategies and improve agricultural practices globally.Graphical AbstractCoastal groundwater supports agriculture, yet escalating salinity degrades both water quality and soil viability. Focusing on Tunisia's Mornag Plain, this study integrates statistical methods (PCA, clustering), GIS, and machine learning (ML) to evaluate irrigation water quality. Critical indices - IWQI, PS, Na%, and SAR - revealed three predominant water types (Na-Cl, Ca-Mg-SO4, Ca-Mg-Cl/SO4) shaped by both natural processes and human activities. Alarmingly, 65% of samples were unsuitable for irrigation, presenting serious risks to agricultural sustainability. The research developed an optimized decision tree model that accurately predicts water quality using minimal parameters, providing a practical and economical assessment tool. This ML approach demonstrates superior efficiency compared to traditional methods, enabling rapid water quality evaluation. The findings offer valuable guidance for arid region water management, supporting evidence-based policy decisions for sustainable irrigation. With global applicability, particularly in coastal agricultural regions, this methodology enables better resource management while promoting environmentally resilient farming practices. The study establishes a framework for addressing groundwater quality challenges through advanced analytical techniques, contributing to long-term food security and ecosystem preservation.