HFLAD: AI-Driven Hierarchical Fusion Learning for Predictive Anomaly Detection in Consumer-Centric Systems

Audit logs, crucial for service auditing and security, capture user and system activities, which can be effectively represented as multivariate time series data to enable efficient log analysis, such as behavior analytics for anomaly detection. However, it is technically challenging to analyze such data due to its high dimensionality and strong time correlation. To address these challenges and enhance predictive security for consumer applications, we propose an AI-driven anomaly detection model named HFLAD. This model comprehensively considers the feature and time correlation of multivariate time series data. It employs causal and dilated convolutions for encoding to capture multi-scale temporal dependencies, and utilizes an HVAE (Hierarchical Variational Autoencoder) generator for information reconstruction. Then, it identifies anomalies by leveraging the difference between the reconstructed data and the original data. By leveraging hierarchical fusion learning, HFLAD proactively detects deviations in consumer-related system activities and audit logs, aligning with modern security demands in smart devices, IoT, and online services. Experimental results demonstrate that HFLAD outperforms other existing techniques in identifying anomalies, thereby improving the effectiveness of predictive security measures for consumer systems.