The increasing demand for water has made rainwater harvesting a crucial strategy to supplement existing surface and groundwater resources, particularly in semi-arid regions. This study evaluates potential surface rainwater harvesting (SRWH) zones in the upper Dwarakeshwar River basin using machine learning methods, including Support Vector Machines (SVM), Random Forests (RF), and Neural Networks (NN). Twelve thematic layers were integrated to develop SRWH suitability maps, categorising the basin into four classes: unsuitable, less suitable, moderately suitable, and most suitable. SVM exhibited the highest classification accuracy (AUC-ROC: 0.815), followed by RF (0.812) and NN (0.75). The SVM model identified 17.62% of the basin as unsuitable and 24.62% as most suitable, resulting in a more balanced and reliable classification than the RF and NN models. Highly suitable zones are primarily associated with gentle slopes, impermeable geology, and adequate precipitation, while Pearson correlation analysis highlights land use and land cover (0.72), drainage density (0.52), and lithology (0.27) as dominant controlling factors. The study also proposes strategic implementation of SRWH structures, including check dams, earthen dams, percolation tanks, farm ponds, and gully plugs, to enhance water retention and groundwater recharge. Validation of predictive models confirmed that SVM was the most effective approach, providing a reliable framework for sustainable water resource management. The findings provide a scalable, cost-effective, and data-efficient decision-support framework for SRWH planning, with future scope for incorporating climate variability to improve water security in arid and semi-arid regions.