A Hybrid Masked MLP for Cyber Intrusion Detection in Sensor Network Domains

Abstract

Cyber intrusion detection in sensor network domains is quite complicated, mainly because of high, dimensional traffic features, redundant data, and strict resource limitations. To overcome these problems, a hybrid intrusion detection system incorporating Principal Component Analysis (PCA), Fuzzy C-Means (FCM) clustering, and a Multilayer Perceptron (MLP) classifier was introduced in this study for efficient and accurate threat detection. While PCA was used to decrease the dimensionality of features and retain the most informative network characteristics, FCM clustering was focused on providing cluster-based membership features that reveal the inner pattern of the traffic. These dimensionally reduced features and the membership features were combined to create a vector of enhanced features and then used in training the MLP model for the binary intrusion classification task. A masking mechanism is placed in the input layer of the MLP which emphasizes intrusion related features and de-emphasizes redundant or noisy features to improve feature learning and reduce computational cost. Various standard performance metrics thereby accuracy, precision, recall, and F1, score were employed to assess the proposed method and achieved the accuracy of 97.24%. Besides, the hybrid masked MLP turns out to be an effective and dependable solution to the problem of cyber intrusion detection in sensor network domains.