Using Deep Learning For Assessing Cybersecurity Economic Risks In Virtual Power Plants

This paper presents an assessment procedure for evaluating economic risks in Virtual Power Plants (VPPs) using two deep learning techniques, viz., Naïve Bayes algorithm and J48 bagging tree model. A comprehensive matrix of cybersecurity economic assessment based on threats, motives, impacts, challenges and economic impacts is provided in this paper. Cybersecurity risk quantification and their measurement and Cyber risk mitigation frameworks are also dealt with in this paper. Our cybersecurity risk Return On Security Investment (ROSI) model employs 9-parameters, while our dataset collection is classified into 11 types of attack profiles. The results obtained from Naïve Bayes classifier and CRQ using J48 bagging tree model yield a predictor accuracy of 82%, while the classifier correctly classifies instances in Naïve Bayes with a Receiver-Operator-Characteristic (ROC) area of 0.940 indicating the good classification of the sampled data; the prediction rate is 94.2% for the CRQ-J48. The ROC curve for the naïve Bayes classifier performs its best when the data is at 80% and performs worse when at 10%. Similarly, the J48 performs best when the data is at 85%-90% and worse between 5%-10%. While examining the cost of the data breach of a single instance of compromise, it can be estimated that web server has the maximum exposure and hence the cost of a breach is approximately estimated to be around 19 million dollars in our dataset.