Crop water stress index (CWSI) is a reliable, economic and non-destructive method of monitoring the onset of water stress for irrigation scheduling purposes. Its application, however, is limited due to the need of obtaining the baseline canopy temperatures. This study developed a self-organizing map (SOM) based model to predict the CWSI using microclimatic variables, namely air temperature, canopy temperature and relative humidity. The canopy temperature measurements were made from Indian mustard crop grown in a humid sub-tropical agro-climate during the 2017 and 2018 cropping seasons. Eight levels of irrigation treatments (I1 – I8) based on maximum allowable depletion of available soil water were considered in the study. The CWSI for treatments I2 – I7 was computed using the empirical approach based on the experimentally measured baseline canopy temperatures from treatments I1 and I8. The number of data points used was 1260 and 1350 for model training and testing, respectively. The developed SOM model was evaluated using the error indices Nash-Sutcliffe efficiency (NSE), bias error (BE), absolute error (AE), and coefficient of determination (R2). The SOM predicted CWSI presented a good agreement with the baseline computed CWSI values during model training (R2 = 0.98, NSE = 0.97, AE = 0.018, BE = 0.0004) and testing (R2 = 0.98, NSE = 0.98, AE = 0.018, BE = 0.002). Treatment specific analysis was conducted to evaluate the performance of SOM predicted CWSI for different irrigation levels. Results indicated that the presence of zero CWSI values in a significant proportion in the dataset impacted the model prediction performance at low CWSI (<0.1) values, with an R2 of 0.71 during testing. Nonetheless, the model performed exceptionally well in predicting CWSI values between 0.1 and 0.6 (R2 = 0.93–0.98, NSE = 0.92–0.98, AE = 0.013–0.015, BE = −0.002–0.004), which is the commonly observed CWSI range for irrigation scheduling in field crops. For better understanding, the developed SOM model was also analysed through the component planes, U-matrix, clusters and high-low bar planes in the cluster features.