Leveraging Federated Learning for File Fragments Classification Based on Depthwise Separable Convolutions

Leveraging Federated Learning for File Fragments Classification Based on Depthwise Separable Convolutions

연구 분야: Safety

학회: ICFNDS '23: Proceedings of the 7th International Conference on Future Networks and Distributed Systems

In digital forensics, file fragment classification plays a crucial role in the file carving process. Recently, convolutional neural network based models have been utilized for file fragment classification to improve the classification accuracy. However, training CNN models requires a large dataset, presenting a challenge in digital forensics where data is sensitive and confidential. To this end, we propose a federated learning framework for file fragments classification based on depth-wise separable convolutions. Accordingly, we can develop a file fragment classification model that is both privacy-preserving and computationally efficient. We experimentally tested the proposed framework using FFT-75 dataset. The experimental results show that the proposed framework achieves comparable accuracies to those of centralized training models while preserving the privacy and confidentiality of sensitive data.