The Insider Threat Detection and Secure Data Transfer Leveraging Bidirectional LSTM with Grouped Orthogonal Initialization and Swish Activation

Threat Detection and Secure Data Transfer

Abstract

Insider threats pose a significant risk to enterprises due to their authorized access to vital systems. This research proposes a novel approach utilizing a Bidirectional Long Short-Term Memory (BiLSTM) model enhanced with Grouped Orthogonal Initialization (GOI) and the Swish Activation function for insider threat detection and secure data transfer. The model incorporates homomorphic encryption for safe data transport and achieves high performance in detecting insider threats with accuracy (0.95), precision (0.92), and AUC-ROC (0.94). The results demonstrate that this model outperforms conventional techniques, providing enhanced security and reliability in insider threat detection.

OBJECTIVES: The primary objectives of this study are to develop an effective method for detecting insider threats with high accuracy and precision, and to ensure the secure transfer of data through encryption techniques. The model also aims to outperform conventional methods by improving security and reducing false positives.

METHODS: The proposed solution utilizes a Bidirectional Long Short-Term Memory (BiLSTM) model, improved with Grouped Orthogonal Initialization (GOI) and the Swish Activation function. Homomorphic encryption is applied to secure data transfers, and the model is trained on insider threat datasets to optimize detection performance.

RESULTS: The model demonstrates exceptional performance in detecting insider threats, achieving an accuracy of 0.95, precision of 0.92, and an AUC-ROC score of 0.94. The results show a significant improvement over conventional methods in both accuracy and security.

CONCLUSION: The proposed BiLSTM model, enhanced with GOI and Swish Activation, provides an effective and reliable solution for insider threat detection. It ensures secure data transfer through homomorphic encryption and outperforms traditional approaches, offering superior accuracy, precision, and security for enterprise systems.