Ensemble LNN for Improved Internet of Things Intrusion Detection

Intrusion detection in networks is a critical area that needs to be taken care of and widely explored by researchers to provide network security in IoT. Installing Network Intrusion Detection (NID) on devices with limited resources remains difficult despite massive research efforts and state-of-the-art technologies. A low-cost intrusion detection technique based on extraction knowledge that balances accuracy and economy by cutting down on model storage and processing expenses concurrently for the Internet of Things is suggested by using an ensemble model by incorporating Lightweight neural networks and XGBoost associated with Principal component analysis (PCA) for feature reduction. In the LNN stacked DeepMax blocks were used to extract compressed representation with XGBoost classifier. The suggested model also handles the performance decline of the LNN by the batch-wise knowledge of self-distillation to regularize training consistency. The observed results of the proposed ensemble model perform better with fewer parameters and lower compute costs.Use of XGBoost gave accuracy of 100%.