Hybrid optimal intrusion detection model for WSN with PCA-based selected feature set.

A complicated problem in Wireless Sensor Network environments is identifying the anomalies since the security risks become more diverse. WSNs have several obstacles, including a shortage of energy, inadequate memory, low computation capacity, and a brief contact range. As a result, improving the recognition precision and convergence speed of intrusion detection in such settings is critical. This paper proposes a hybrid optimal intrusion detection model for WSN to enhance detection accuracy. The data with class imbalance problem is handled with SMOTE method. Raw features, higher-order statistical, and entropy-based features are the set of features extracted from the balanced data. The extracted features are subjected to the PCA-based feature selection process for selecting the relevant data from the extracted features to make the process more accurate. A hybrid model (combining QNN and Deep Maxout) trains the selected features to decide on the presence of intrusion. In this model, the optimal training is done with the hybrid RUJSO algorithm that combines the algorithms like ROA and JFO, through which the optimal weights of the hybrid model are tuned. The attacker node is mitigated from the network once the model identifies the attack. [ABSTRACT FROM AUTHOR]