A feature-enhanced FE-LDA-based method for syslog detection

A significant amount of log data is generated during system operation, which is extensively used for system maintenance and troubleshooting. With the widespread adoption of deep learning, sophisticated log detection models are increasingly employed for system log analysis. However, deep learning-based log detection models are highly dependent on training data, have complex architectures, and require numerous hyperparameters, resulting in inefficiency and excessive computation times that hinder timely detection of abnormal logs. Compared to deep learning algorithms, enhanced traditional machine learning methods are more favorable for log detection. In this study, we propose a system log detection method based on feature-enhanced FE-LDA. First, raw log information is parsed using templates through log parsing. The parsed logs are then grouped and feature-enhanced using multi-granularity scanning to extract log feature vectors. Finally, an enhanced linear discriminant analysis (LDA) algorithm is applied for log fault detection, providing fault identification results. Experimental results demonstrate that the proposed model outperforms deep learning-based log detection models in terms of both effectiveness and stability.