Contrastive Learning-Based Anomaly Detection for Actual Corporate Environments.

Area of Science:

• Computer Science
• Information Systems
• Machine Learning

Background:

• Information systems are critical for business operations, including personnel, budget, and financial management.
• System anomalies can cause complete operational paralysis, necessitating robust detection methods.
• Collecting and labeling anomalous data from live corporate systems presents significant challenges due to stability constraints and data imbalance.

Purpose of the Study:

• To propose a deep learning method for anomaly detection in corporate information systems.
• To address the challenges of small and imbalanced datasets in real-world operating environments.
• To develop a method suitable for environments where collecting anomalous data is difficult.

Main Methods:

• Utilized contrastive learning with data augmentation through negative sampling.
• Focused on creating and labeling datasets from actual corporate operating systems.
• Compared the proposed method against traditional deep learning models like Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM).

Main Results:

• The proposed method achieved a true positive rate (TPR) of 99.47%.
• CNN and LSTM models achieved lower TPRs of 98.8% and 98.67%, respectively.
• Demonstrated superior effectiveness in anomaly detection, particularly for small datasets.

Conclusions:

• The proposed contrastive learning method is highly effective for anomaly detection in corporate information systems.
• The technique is particularly advantageous when dealing with small or imbalanced datasets.
• This approach offers a viable solution for maintaining the stability and integrity of critical business systems.