This study aimed to correlate the physicochemical attributes of rice grains using non-destructive techniques combined with machine learning algorithms. Samples of white, black, red, and parboiled rice were analyzed using hyperspectral spectroscopy (350-2 500 nm) and subjected to linear regression (LR), random forest (RF), gradient boosting (GB), support vector machine (SVM), convolutional neural networks (CNN), and recurrent neural networks (RNN) prediction. Spectral and physicochemical data were examined through multivariate analysis (PCA) and cross-validation using performance metrics such as R, R², MAE, and RMSE. The results indicated that SVM, RF, and GB outperformed the other algorithms, showing higher accuracy and lower variability in the prediction, SVM reached R = 0.952 and R² = 0.904, with MAE = 0.409 and RMSE = 0.582, followed closely by RF with R = 0.950 and MAE = 0.416, and GB with R = 0.947 and MAE = 0.431. Black rice stood out for its high protein, lipid, and ash contents. Parboiled rice showed higher fiber content, and white rice was notable for its elevated starch content. Hyperspectral spectroscopy proved effective in differentiating rice types, enabling the identification of relevant spectral bands for optimized sensors. Overall, integrating non-destructive technologies with machine learning shows strong potential for industrial applications.