Hybrid Model for Intrusion Detection in Wireless Sensor Network: An Improved Class Imbalance Processing

Abstract

A significant difficulty in WSN settings is recognizing the abnormalities as security threats become divergent in various fields. The major drawbacks of WSN including insufficient memory, limited energy, and low compute power, and a small communication range. Thus, enhancing the detection accuracy of intrusion detection in such contexts is critical. However, this work intends to propose intrusion detection in WSN with improved class imbalance processing. The input data is pre-processed to balance the data with modified class imbalance process. Here, the SMOTE-ENN and Tomek link algorithm is employed to pre-process the raw data. Then the entropy and improved correlation based features are retrieved from the balanced data. Later, these features are trained by subjecting those features into the hybrid model that includes Deep Maxout and Bi-GRU model and then the final detection is predicted with the classifier outcomes. Further, at the training rate 90%, the proposed yielded the least FPR rate (0.1038) than the other 60, 70 and 80 training percentages.