The Crop Water Stress Index (CWSI), a pivotal indicator derived from canopy temperature, plays a crucial role in irrigation scheduling for water conservation in agriculture. This study focuses on determining CWSI (by empirical method) for wheat crops in the semi-arid region of western Uttar Pradesh, India, subjected to varying irrigation treatments across two cropping seasons (2021-2022 and 2022-2023). The aim is to investigate further the potential of four machine learning (ML) models-support vector regression (SVR), random forest regression (RFR), artificial neural network (ANN), and multiple linear regression (MLR) to predict CWSI. The ML models were assessed based on determination coefficient (R2), mean absolute error (MAE), and root mean square error (RMSE) under diverse scenarios created from eight distinct input combinations of six variables: air temperature (Ta), canopy temperature (Tc), vapor pressure deficit (VPD), net solar radiation (Rn), wind speed (U), and soil moisture depletion (SD). SVR emerges as the top-performing model, showcasing superior results over ANN, RFR, and MLR. The most effective input combination for SVR includes Tc, Ta, VPD, Rn, and U (R2 = 0.997, MAE = 0.901%, RMSE = 2.223%). Meanwhile, both ANN and MLR achieve optimal results with input combinations involving Tc, Ta, VPD, Rn, U, and SD (R2 = 0.992, MAE = 2.031%, RMSE = 3.705%; R2 = 0.759, MAE = 13.95%, RMSE = 19.98%, respectively). For RFR, the ideal input combination comprises Tc, Ta, VPD, and U (R2 = 0.951, MAE = 5.023%, RMSE = 9.012%). The study highlights the considerable promise of ML models in predicting CWSI, proposing their future application in integration into an irrigation decision support system (IDSS) for crop stress mitigation and efficient water management in agriculture.