A New Deep Learning Approach Enhanced with Ensemble Learning for Accurate Intrusion Detection in IOT Networks

The size and capabilities of IoT (Internet of Things) networks have seen an unprecedented growth in recent years. Consequently, many organizations have deployed large scale IoT networks to increase their organizational efficiency. However, the added benefits of IoT networks come along with a higher risk of malicious attacks and intrusions. Robust and accurate IDSs (Intrusion Detection Systems) are hence vital in preventing damage and taking preventive measures. Until recently, IDSs were created using conven-tional machine learning algorithms such as SVM, decision trees and random forests. Although these algorithms provided decent results, the systems created were inflexible and non-scalable. In contrast, deep learning methods have been shown to perform considerably better in situations where complex relationships exist within the data. Additionally, other approaches such as ensemble learning provide an opportunity to improve the accuracy of the results as well as develop a scalable distributed system. In this paper, we present a methodology to create efficient IDS combining the strengths of deep learning and ensemble learning. Utilizing these approaches, an ensemble of Feedforward Neural Networks (FNN) is created to detect intrusions and pre-vent attacks. The performance of the approach is validated using k-fold cross validation on a sample from the Bot-IoT dataset. Furthermore, a comparison is done with Random Forest, Decision Tree and Xgboost models to see the efficacy of the approach. Results obtained from the k-fold cross validation of the deep ensemble approach show a high classification accuracy of 99.08% on the Bot-IoT dataset.