Satellite image-based remote sensing is an effective and non-destructive approach for monitoring agricultural crop performance and can play an important role in resource management, supply chain planning, and food security assurance. Integrating remote sensing data with data mining methods and machine learning algorithms can further improve the accuracy of crop yield prediction. This study was conducted to develop and evaluate an analytical framework based on satellite data and machine learning algorithms for estimating cotton yield in Golestan Province. Considering the economic importance of cotton and the need for timely data in agricultural management, Landsat 8 and 9 satellite images were utilized. After extracting spectral and vegetation indices, the performance of Multiple Linear Regression (MLR) and the M5 decision tree model was evaluated using error metrics. The results indicated that the MLR model had lower accuracy compared to the M5 model. The best results were obtained on October 8 (October 16, 1403 in the Iranian calendar), coinciding with the peak flowering and boll formation stage of cotton. Specifically, the M5 model achieved the highest predictive performance on this date, with a coefficient of determination (R²) of 0.75, a Root Mean Square Error (RMSE) of 846 kg ha⁻¹, and a Mean Absolute Error (MAE) of 671 kg ha⁻¹. In contrast, the MLR model yielded an R² of 0.46, an RMSE of 1,207 kg ha⁻¹, and an MAE of 1,026 kg ha⁻¹.. Among the evaluated indices, SAVI, NDVI, EVI, LAI, and NMDI showed the greatest contribution to yield prediction, highlighting the significant role of vegetation cover, leaf area, and plant moisture status in determining final yield. The findings demonstrated that the M5 algorithm is an accurate, cost-effective, and interpretable method for regional cotton monitoring and for developing intelligent precision agriculture systems in Iran.