<p>Rice yield prediction before harvest enables farmers in optimizing the risk of yield loss and adoption of suitable agronomic management practices. Remote sensing data captured using low cost RGB sensors mounted on drones offers a reliable alternative to conventional methods. This study was conducted at ICAR-Central Rice Research Institute, Cuttack, Odisha, India using different rice cultivars, with the objective of developing machine learning (ML) models for rice yield prediction using predictor variables (vegetation indices) generated from data captured using RGB sensors. From 25 vegetation indices (VIs) derived during the panicle initiation stage via RGB imagery, 17 were selected for model development using the variance inflation factor (VIF) technique. Using the selected vegetation indices as a predictor variable, four machine learning models gradient boosting machine (GBM), random forest (RF), artificial neural network (ANN), and support vector machine (SVM) were developed with rice yield as the dependent variable. Among all model, the RF model achieved best prediction accuracy in predicting rice yield (RMSE = 0.27 t ha⁻¹). The RF model predicted yield in the range of 3.16–5.50 (t ha⁻¹). The coloration (CI), hue (H) and intensity (I) vegetation indices were identified as the most important contributors in accurate prediction of rice yield across all the ML models. The study demonstrated that the proposed approach is reliable and cost-effective method of the of rice grain yield prediction however, models need multi-season and multilocational validation before suggesting a large-scale application.</p>