This study presents a method for classifying apple varieties using deep features and machine learning. The research focuses on using transfer learning with seven popular convolutional neural network (CNN) architectures—VGG16, VGG19, InceptionV3, MobileNet, Xception, ResNet150V2, and DenseNet201—to classify 10 apple varieties. DenseNet201 achieved the highest classification accuracy of 97.48%. Deep features were then extracted using DenseNet201 and used to train traditional machine learning models, including support vector machine (SVM), random forest classifier (RFC), multi-layer perceptron (MLP), and K-nearest neighbor (KNN). The SVM model achieved an accuracy of 98.28%, while the MLP model achieved the highest accuracy of 99.77% after applying principal component analysis (PCA) for dimensionality reduction. The study also evaluated the performance of these models using various metrics, including accuracy, precision, recall, F1-score, and AUC-ROC. The results showed that the MLP model outperformed all other models, achieving the highest accuracy, precision, and recall. The integration of deep features, PCA, and traditional machine learning models proved effective in improving classification accuracy. The study concludes that the proposed method can classify ten apple varieties with high accuracy and is a promising approach for future research in apple variety classification. The results indicate that the integration of deep features, PCA, and machine learning models can enhance the accuracy and performance of apple variety classification. The study also highlights the importance of using transfer learning and dimensionality reduction techniques in improving the performance of machine learning models for apple variety classification.This study presents a method for classifying apple varieties using deep features and machine learning. The research focuses on using transfer learning with seven popular convolutional neural network (CNN) architectures—VGG16, VGG19, InceptionV3, MobileNet, Xception, ResNet150V2, and DenseNet201—to classify 10 apple varieties. DenseNet201 achieved the highest classification accuracy of 97.48%. Deep features were then extracted using DenseNet201 and used to train traditional machine learning models, including support vector machine (SVM), random forest classifier (RFC), multi-layer perceptron (MLP), and K-nearest neighbor (KNN). The SVM model achieved an accuracy of 98.28%, while the MLP model achieved the highest accuracy of 99.77% after applying principal component analysis (PCA) for dimensionality reduction. The study also evaluated the performance of these models using various metrics, including accuracy, precision, recall, F1-score, and AUC-ROC. The results showed that the MLP model outperformed all other models, achieving the highest accuracy, precision, and recall. The integration of deep features, PCA, and traditional machine learning models proved effective in improving classification accuracy. The study concludes that the proposed method can classify ten apple varieties with high accuracy and is a promising approach for future research in apple variety classification. The results indicate that the integration of deep features, PCA, and machine learning models can enhance the accuracy and performance of apple variety classification. The study also highlights the importance of using transfer learning and dimensionality reduction techniques in improving the performance of machine learning models for apple variety classification.