Optimized Feature Selection techniques for Distributed Intrusion Detection System (DIDS) in IoT Environment

Abstract

Introduction: In Internet of Things (IoT), optimal data set management and feature learning are the important problems that affect the attack detection’s accuracy.

Objectives: To select the optimal features in the dataset, using advanced optimization techniques and classify the data as normal or anomaly,using ML based algorithms.

Methods: In this paper, we propose to design anoptimized feature selection technique for Distributed Intrusion Detection System (DIDS)in IoT environments. During the preprocessing phase, t-distributed Stochastic Neighbor Embedding (t-SNE) is applied for data exploration and visualizing the high-dimensional data and Principal Component Analysis (PCA) technique is applied for dimensionality reduction.Then, for selecting the optimal features from the preprocessed dataset, the Improved Gravitational Search Algorithm (IGSA) is applied. Finally, for classifying the data as normal or anomaly, the XGBoost classifier is applied.

Results: Experimental results show that the optimized XGBoost classifier attains highest accuracy and F1-score values, when compared to the other classifiers

Conclusion: The proposed DIDS thus protects the IoT networks from external attacks quickly and effectively