Federated-learning-framework-based malware classification algorithm under insufficient local training data condition
Ruifeng Yu, Jun Yan · 2025
With the development of the Internet, malware classification has become a very important research issue in cybersecurity. Traditional centralized learning based malware classification techniques pay little attention to data privacy protection.In order to solve this issues, in this paper, a federated learning framework based malware classification algorithm under insufficient local training data condition is proposed. First, in order to take advantage of deep learning network, a training data preprocessing is proposed to transform the malware binary code into a gray image. Then, in order to keep the local training performance, the transfer learning technique based coarse learning is proposed to solve the problem of small number of training date. At last, a hierarchical FL based refined learning is proposed to improve malware classification performance by two different FL processing. Experimental results show that the proposed algorithm can obtain good malware classification performance.