Cyber Security Threat Prediction using Time-Series Data With LSTM Algorithms

Abstract

Cyber security remains a paramount concern in the digital era, with organizations and individuals increasingly vulnerable to sophisticated cyber-attacks. This study aims to develop and evaluate Long Short-Term Memory (LSTM) regression models to predict three types of cyber attacks: flood, spyware, and vulnerability. The LSTM algorithm is used to construct regression models for spyware, flood, and vulnerabilities within a firewall log dataset. The experiments demonstrate that preprocessing techniques such as normalization and standardization can positively impact model performance by reducing prediction errors and enhancing accuracy. The results of the experiments show that the model developed in this research exhibits potential in predicting cyber attacks. For the flood attack model, the best performance was achieved with an RMSE of 59.8810 and an R-Squared of 0.9214 after data standardization. The spyware attack model's best results were an RMSE of 133.9567 and an R-Squared of 0.7685 after standardization. In contrast, the vulnerability attack model showed limited improvement, with the best RMSE of 503.5521 and an R-Squared of 0.2358 after standardization. Moreover, real-time implementation and testing of these models in live network environments could validate their practical applicability and effectiveness.