S2SCAF-BNRA: Sequence to Sequence Based Approach for Cyber-Attack Forecasting with Bayesian Network Decision-Making

Cyber-attack forecasting is a proactive way that allows organizations to prevent intrusions and reduce the impact of cyber-attacks. Due to the increasing methods and damages of cyber-attacks, researchers are increasingly interested in proactive detection methods, exploiting advances in the field of AI. In this paper, we propose a proactive approach to reduce cyber-attack risk. The proposed approach consists of two main components: Sequence To Sequence Cyber-Attack Forecasting (S2SCAF), and Bayesian Network Risk Assessment (BNRA). The S2SCAF component is a sequence-to-sequence deep learning-based model that takes a multivariate time series as input to forecast the cyber-attack rate at the next step. The BNRA component is a Bayesian network-based decision-making model, which is used to evaluate the future risk level of the system security situation and alert the security administrator. Using the CICDDoS2019 dataset, we evaluate S2SCAF and compare it against Auto-Regressive, Moving Average, Exponential Smoothing, and CNN-LSTM models. Forecasting results show that the proposed S2SCAF method outperforms the comparison models with a Mean Absolute Error (MAE) of 0.344, a Mean Squared Error (MSE) of 0.381, and Root MSE (RMSE) of 0.617. Moreover, decision-making results show that the BNRA component can capture the causal relationships between the risk factors and the risk level, hence offering a decision-support tool for security administrators.