ADFS: An Attention-Driven Feature Selection Approach for Data Security-Related Traffic Analysis

Abstract

With the exponential growth of digital data across global networks, data security has become a critical concern for organizations worldwide. The threats targeting sensitive data have evolved significantly, with sophisticated attacks capable of bypassing traditional security mechanisms and exfiltrating valuable information. This paper proposes an Attention-Driven Feature Selection (ADFS) approach specifically designed for data security-related traffic analysis, addressing the challenges of detecting subtle data exfiltration patterns in network traffic. Unlike existing methods that rely on manual feature engineering or uniform feature weighting, our approach leverages a multihead attention mechanism to automatically learn feature importance through end-to-end training. This allows the model to identify the most discriminative features for different types of data security threats while reducing computational overhead. We implement a lightweight feature fusion framework that combines statistical and behavioral features, enabling efficient detection of data security anomalies. Comprehensive experiments on the CICIDS2017 dataset demonstrate that our approach achieves superior detection performance (97.8 % accuracy) compared to state-of-the-art methods while reducing feature dimensionality by 63% on average. Furthermore, the attention-based mechanism provides valuable interpretability for security analysts, highlighting the specific traffic characteristics most relevant to potential data security violations. The proposed ADFS framework represents a significant advancement in automated protection against data theft, unauthorized data access, and other data security threats in modern network environments.