Comparison of CNN and HOG-Based Machine Learning Methods for AI-Generated Image Classification Using the CIFAKE Dataset

Dika Kurnia Putra (Faculty of Engineering and Informatics, Univeritas Bina Sarana Informatika, Indonesia)
Chandra Ari Pradana (Faculty of Engineering and Informatics, Univeritas Bina Sarana Informatika, Indonesia)
Muhammad Hilal Gilardin (Faculty of Engineering and Informatics, Univeritas Bina Sarana Informatika, Indonesia)
Jordy Lasmana Putra (Faculty of Engineering and Informatics, Univeritas Bina Sarana Informatika, Indonesia)

Journal of Information System Exploration and Research
Vol. 4 No. 4: October 2026

The rapid development of Generative Artificial Intelligence (Generative AI) has enabled the creation of highly realistic images that are increasingly difficult to distinguish from real images. This raises concerns about misinformation, identity forgery, and digital media manipulation, highlighting the need for reliable image classification methods. This study compares Convolutional Neural Network (CNN) features with Histogram of Oriented Gradients (HOG)-based machine learning methods, including K-Nearest Neighbor (KNN), Decision Tree (DT), and Random Forest (RF), for classifying AI-generated and real images. The CIFAKE dataset, consisting of real and AI-generated images, was used in the experiments. CNN performed automatic feature extraction, while KNN, DT, and RF used HOG feature vectors. Model performance was evaluated using accuracy and confusion matrix analysis. The results show that CNN achieved the highest accuracy of 87.30%, followed by Random Forest (75.00%), KNN (72.27%), and Decision Tree (63.98%). These findings indicate that CNN is more effective in capturing visual patterns in AI-generated images, while HOG-based methods offer lower computational complexity. Therefore, CNN provides better classification performance for AI-generated image detection.