Correlating Test Events With Monitoring Logs For Test Log Reduction And Anomaly Prediction

Abstract

Automated fault identification in long test logs is a tough problem, mainly because of their sequential character and the impossibility of constructing training sets for zero-day faults. To reduce software testers' workload, rule-based approaches have been extensively investigated as solutions for efficiently finding and predicting the fault. Based on software system status monitoring log analysis, we propose a new learning-based technique to automate anomaly detection, correlate test events to anomalies and predict system failures. Since the meaning of fault is not established in system status monitoring-based fault detection, the suggested technique first detects periods of time when a software system status encounters aberrant situations (Bug-Zones). The suggested technique is then tested in a real-time system for anomaly prediction of new tests. The model may be used in two ways. It can assist testers to focus on faulty-like time intervals by reducing the number of test logs. It may also be used to forecast a Bug-Zone in an online system, allowing system administrators to anticipate or even prevent a system failure. An extensive study on a real-world database acquired by a telecommunication operator demonstrates that our approach achieves 71 % accuracy as a Bug-Zones predictor.