ENHANCING RICE QUALITY PREDICTION USING A HYBRID DEEP LEARNING MODEL WITH EXPLAINABLE AI TECHNIQUES

This research explores the use of a hybrid deep model in rice quality prediction using Explainable AI methods like SHAP and Grad-CAM.The model is compared with ResNet50, VGG16, CNN, and MobileNetV2 based on Kaggle's dataset, which includes images of rice grains and their physical and biochemical properties.Experiment results confirm that the hybrid architecture outperforms other architectures on critical performance measures with the best accuracy (0.95) compared to ResNet50 (0.92) over VGG16 and CNN, which are not good.Sensitivity analysis also confirms the high true positive rate of the hybrid model with the least false negatives, while CNN has the poorest sensitivity.In terms of specificity, the suggested model performs the best via the accurate identification of negative cases alongside minimizing false positives.F1-score analysis testifies to its strength in the trade-off between recall and precision concerning existing architectures.It enhances the classification accuracy and extraction of features based on multiple means of learning and thus makes for more reliable prediction.The