Privacy Preserving Approach for Detection of Malicious Files

With the growing need for malware detection, deep learning models have proven instrumental in identifying and mitigating malware, particularly when tailored to a single organizational network. However, owing to privacy and security concerns, there is a need to develop a distributed malware detection system in which training data from diversified environments can be utilized for detection. The emerging need for cybersecurity due to the increase in cyberattacks has increased the demand for privacy-supporting technologies. In this study, we propose a privacy-preserving distributed malware detection model that employs the Federated Credibility Index-based Aggregation model (FedCIA). The deep MLP architecture processes the features of the malware to perform local training and generate Local Parameter Updates (LPU). These parameters are subsequently sent to the server, where the Credibility Index is calculated for each client. Based on Credibility, the server uses the client’s parameters for the aggregation process, with which Global Parameter Updates(GPU) are generated. The GPU is then redistributed to all the clients connected to the server to revise their local models. The experiments conducted using this proposed model achieved competitive performance and addressed the challenges associated with unknown intermittent client participation, promoting cooperative efforts among cybersecurity organizations to improve diverse individual datasets for effective malware detection.