Evaluation of Federated Learning Based on Malware Detection

FL-LR, or Federated Learning Logistic Regression, enhances cybersecurity by using Federated Learning to detect malware in executable files. This approach allows decentralized devices to collaborate on training a Logistic Regression model without sharing raw data directly. Each device holds part of the malware dataset, ensuring data confidentiality. FL-LR's effectiveness is compared with Support Vector Machines and Random Forests using metrics like accuracy, precision, memory usage, F1-score, and AUROC. Results show FL-LR performs well while preserving data privacy. The study also examines FL-LR's scalability and convergence in federated settings, demonstrating its suitability for large-scale malware detection. Overall, FL-LR with Logistic Regression shows promise for enhancing cybersecurity through Federated Learning while protecting sensitive data.