This research work develops a hybrid approach for rice grain image classification by employing MobileNetV2 pretrained model for feature extraction, Principal Component Analysis (PCA) for dimensionality reduction and supervised machine learning classifiers for classification. The rice grain dataset consists of 75,000 images divided into five different classes, with equal number of images for each. Experimental results shows SVM achieved highest accuracy of 95.79% followed closely by Catboost ensemble model of 95.17% accuracy. PCA was employed to reduce dimension while retaining significant features and reducing computational complexity. This research discovers the likely advantages of the integration of deep learning and machine learning approaches for efficient and scalable image classification, which is vital in applications such as agriculture, food quality control and other related fields.