A Deep Autoencoder Based Outlier Detection Model for Intrusion Detection Systems in Wireless Sensor Networks

With advancements in Wireless Sensor Network technologies, a large inflow of data is observed in the networks. Security threats have been rising in the recent years. These frequent security threats have created an urgency to upgrade the Intrusion Detection Systems to highly complex methods. Hence, increasing the computation overhead on the energy limited WSN nodes. These nodes sometimes undergo data or network anomalies while transmitting or receiving packets, creating outliers. In order to remove the outlier, a Hybrid Convolutional Neural Network, Fully Connected Network based Autoencoder has been proposed. The proposed algorithm successfully mitigated more than 31,000 outlier instances. Further, Random Forest classifier gave a high accuracy of 99.975% on the reduced dataset.