Cybersecurity Risk Prediction Using Fuzzy Logic based Models

In modern digital security, predicting cyber security risk is a critical aspect. This research is presenting a comparative analysis of two distinct AI (Artificial Intelligence) based models viz. Fuzzy Inference System and Adaptive Neuro Fuzzy Inference System for predicting cyber security risk on the basis of 3 parameters namely authentication strength, traffic anomaly and network load. Both the models are developed in MATLAB, for which a data set of 100 samples is utilized, 85% for training and 15% for testing. Gaussian membership functions have been formed for all input and output variables. The evaluation of this system is done comparing the anticipated risk values with actual values. The ANFIS model has been initialized with grid partition fuzzy inference system. It is trained in 200 epochs considering a small tolerance for ensuring convergence. This trained model is hence tested on the unseen inputs for predicting cyber security risk. Its output has been compared with that of fuzzy model. Evaluation metric namely root mean square error (RMSE) has been used for assessing the prediction accuracy. It has been found that the ANFIS is outperforming the conventional FIS approach displaying a smaller R.M.S.E. and enhancing prediction accuracy. This study outlines the advantages of integrating neuro adaptive learning into FIS for assessing cyber security risk, upgrading risk prediction accuracy. Therefore, ANFIS comes out to be more reliable technique pertaining to cyber security application.