Insider Attack Detection Using Deep Learning

Abstract

In today’s world the most dangerous security threats are not launched by malicious outsiders or malware but from trusted insiders. The exploitation and leakage of sensitive data and information by malicious insiders is getting worse day by day. According to “Insider Report 2018” 90% of the organizations are prone to insider attack. Around 33% organizations encountered insider attacks in the last 12 months. However, most recent advancements and research in this field focus on using machine learning techniques for the detection of insider attacks because clues of malicious behavior of an employee may be extended over multiple datasets, concealed among hundreds of thousands of other data points, or mixed with normal user behavior, or separated by weeks or months of idleness. Now a days deep learning is a trending research topic and it is being applied in various security frameworks due to its enormous advantages. Deep learning has enormous advantages. The algorithms outperform traditional machine learning algorithms in both performance and accuracy. However, there is no prior in-depth research in the field of insider attack detection. Insider attack detection using deep learning is still an open challenge. So the purpose of our research is the detection of insider threat by proposing a deep learning based novel approach LSTM-AutoEncoder. The proposed approach is compared to other techniques in terms of Accuracy, Precison and F1 Score and produces significant results.