EHML: An Efficient Hybrid Machine Learning Model for Cyber Threat Forecasting in CPS

Abstract

Cyber-attacks have risen to prominence as the world’s most serious issue. The rise in cyber-attacks also introduces several new cyber crimes. Cyber-attack detection and prevention are challenging jobs. However, several investigators have developed Artificial intelligence-based cybercrime detection and prevention security models in their past research. Machine Learning (ML) methods play a vital role in cybercrime detection and classification. The most challenging part of machine learning techniques is identifying a suitable model to describe the training data set. This research presents an efficient hybrid ML model (EHML) for cybercrime detection and classification. The proposed EHML model utilizes supervised learning for crime detection and unsupervised learning-based data reduction technique to construct the perfect cyber crime analysis model. The proposed EHML model selects all the relevant and essential features by utilizing an enhanced decision tree (EDT) strategy and Enhanced Local Outlier Factor (ELOF) Technique. The EDT method is based on an enhanced feature selection concept that recursively eliminates all the essential characteristics of the dataset. The ELOF Technique helps to determine all the anomalies and outlier data from the dataset. This research utilizes Kaggle online cybercrime dataset to verify the performance of proposed EHML and existing ML methods, i.e., SVM, J-48, and Random forest. Experimental findings demonstrate that the proposed EHL model obtains an accuracy rate of 95.02%, a precision of 95.01%, a recall of 94.89%, and an F1-score of 95.89 %, which is 10 % better than existing ML methods.