FILE TYPE CLASSIFICATION USING MACHINE LEARNING

Abstract

File classification is crucial in information security, system management, and digital forensics. Traditional methods like classification by file extensions, header extraction, or basic machine learning have limitations such as low accuracy and poor scalability. This thesis proposes a file classification approach using byte histograms combined with Decision Tree and Random Forest models, enhanced by supplemental features like entropy and file size. A dataset with 12 common file types was used, with 80% for training and 20% for testing. The Random Forest model with additional features achieved the highest accuracy of 92.5%, outperforming Decision Tree in Precision, Recall, and F1-Score, especially for types like pdf, exe, and json. The proposed method offers high accuracy, scalability, and practical applicability for file classification tasks.