Optimal Hybrid Deep Belief Networks based Cyberattack Detection in Blockchain Assisted Internet of Things Environment

In recent years, the Cyber Attack Detection (CAD) is the major action for detecting and responding to the malicious attacks or illegal activities in networks or in digital environments. In existing, the vulnerable threats are detected with the help of several Deep Learning (DL) techniques including Convolution Neural Networks (CNN). However, the existing models are inadequate and vulnerable to the intrusions. In this research, Hybrid Deep Belief Networks (HDBN) is proposed to reduce the impact of vulnerable intrusions. Initially, the data is considered from Network Security Laboratory - Knowledge Discovery and Data Mining (NSLKDD). Then, the data is further preprocessed with min-max scaling to transform higher dimensional data into lower dimensional data. After that, Recursive Feature Elimination in combination with Random Forest (RFE-RF) is introduced to select the features from preprocessed data based on importance score. Finally, the detection of cyberattacks is done through proposed HDBN by learning complex patterns in a network data. The proposed HDBN reduced the false negative and provided better results including accuracy (99.20%), precision (96.50%), recall (80.52%), F1-score (80.23%) and Area Under Curve (AUC) (90.36%) when compared to existing Honey Badger Algorithm.