LogABC: A Log Anomaly Detection Model Based on Deep Learning

Logs record important status information during system operation, and automated log anomaly detection can accurately locate the cause of system failures. However, log statements can evolve over time and have instability issues. Therefore, a new method for anomaly detection of log sequences is proposed. Using Bidirectional Encoder Representation from Transformers model and quantity feature extraction model to extract log semantic features and quantity features, then using Convolutional Neural Network to extract local information of log sequence semantic features, using Bidirectional Gate Recurrent Unit to extract contextual information of log sequence semantic features, and using Adaptive Boosting algorithm to fuse quantity features and local and contextual information of semantic features for anomaly detection. The performance of the proposed method was validated on HDFS and BGL datasets. Compared with 6 benchmark models, this model has the best precision and F1-score. The experimental results indicate that this method can effectively handle the problem of data instability in log sequences.