Privacy-Preserving Federated Learning in Network Intrusion Detection: A Systematic Literature Review

Abstract

Machine learning privacy preservation is essential because it defends against misuse and illegal access to sensitive personal data including financial information, medical records, and behavioral patterns. Centralizing data in one place is necessary for traditional machine learning techniques, which poses serious privacy problems. Federated learning becomes an innovative approach in this situation. With federated learning, the model comes to the data rather than the other way around, radically altering the training process for machine learning systems. Individual devices or organizations where the data is naturally located are used for the training process rather than a central server. Each participant trains the model using local data, and only model updates are returned to the global model for updating. Raw data never leaves its original place, hence there is a far lower chance of data breaches during transfer. This article proposes a systematic review of federated learning with privacy preservation for intrusion detection. The three chosen online libraries of IEEE Xplorer, Scopus, and Web of Science are searched. Each database has its corresponding search query. The search is conducted to include all papers published since 2016 on computer science research area. The search results contain 220 papers from the different search engines. After removing duplicated papers, the search results are reduced to 131. Inclusion and exclusion criteria are then used to filter the search results. After applying the criteria of inclusion and exclusion, only 32 papers are approved.