Efficient Data Noise-Reduction for Cyber Threat Intelligence System

Preemptive respondents on cyber threats have become an essential part of cybersecurity. Cyber Threat Intelligence (CTI) is an evidence-based threat detection and prevention system. CTI system analyzes and shares the security data to mitigate evolving cyber threats using security-related data. However, to gather enough amount of data for analysis, the CTI system uses various data collection channels. The reliability of data collected from these channels is a critical issue because the inaccurate and vast amount of information could degrade the performance of threat detection. Thus, proper filtering is needed to remove the noise data. In this paper, we propose a data noise-reduction algorithm. The proposed algorithm reflects the contextual characteristics of CTI data and reduces noise data in the CTI dataset. Noise-reduced dataset increases the performance of machine learning and deep learning-based attack prediction models. In our experiment, we conducted a cyber-attack classification using a noise-reduced CTI dataset. As a result, we improve the accuracy of classification from 84 to 96% and reduce the volume of the dataset by 70%.