A Privacy-Preserving Distributed and Incremental Learning Method for Intrusion Detection

Computer systems are facing an increased number of security threats, specially regarding Intrusion detection (ID). From the point of view of Machine learning, ID presents many of the new cutting-edge challenges: tackle with massive databases, distributed learning and privacy-preserving classification. In this work, a new approach for ID capable of dealing with these problems is presented using the KDDCup99 dataset as a benchmark, where data have to be classified to detect an attack. The method uses Artificial Neural Networks with incremental learning capability, Genetic Algorithms and a feature selection method to determine relevant inputs. As supported by the experimental results, this method is able to rapidly obtain an accurate model based on the information of distributed databases without exchanging any compromised data, obtaining similar results compared with other authors but offering features that make the proposed approach more suitable for an ID application.