FEDetect: A Federated Learning-Based Malware Detection and Classification Using Deep Neural Network Algorithms

Abstract

The growing importance of data security in modern information systems extends beyond the preventing malicious software and includes the critical topic of data privacy. Centralized data processing in traditional machine learning methods presents significant challenges, including greater risk of data breaches and attacks on centralized systems. This study addresses the critical issue of maintaining data privacy while obtaining effective malware detection and classification. The motivation stems from the growing requirement for robust and privacy-preserving machine learning methodologies in response to rising threats to centralized data systems. Federated learning offers a novel solution that eliminates the requirement for centralized data collecting while preserving privacy. In this paper, we investigate the performance of federated learning-based models and compare them classic non-federated approaches. Using the CIC-MalMem-2022 dataset, we built 22 models with feedforward neural networks and long short-term memory methods, including four non-federated models. The results show that federated learning performed outstanding performance with an accuracy of 0.999 in binary classification and 0.845 in multiclass classification, despite different numbers of users. This study contributes significantly to understanding the practical implementation and impact of federated learning. By examining the impact of various factors on classification performance, we highlight the potential of federated learning as a privacy-preserving alternative to centralized machine learning methods, filling a major gap in the field of secure data processing.