Diagnosing Software Bugs using AI-based Root Cause Analysis of Log Data

Faults in software systems predispose them to exhibit behaviors that can be used to fingerprint them. These fingerprints, captured as chronological snapshots in a system’s persistent runtime logs, carry information that can potentially aid in pinpointing the originating location of the fault. Thus, manual or automatic analysis of these captured snapshots is utilized in determining the health status of the system, whether nominal or anomalous. A system in a non-nominal state requires root-cause analysis of the faults’ symptoms encapsulated in the anomalous log data. However, the complexity prevalent in today’s software systems invalidates the manual approach of troubleshooting faults from log data and necessitates powerful DL-based automatic techniques. As such, this research pursues the log data-based defect diagnosis tasks of defect classification and defect localization, leveraging LSTM, Bi-LSTM, and Transformer models. While the developed defect classification Bi-LSTM and Transformer models were adept at categorizing defects, they also provided insights useful for understanding the relationship between the utilized Hadoop Distributed File System (HDFS) log data processing method and their prediction accuracy. To localize bugs, a summarization encoder-decoder LSTM and Transformer models were trained by establishing a novel technique denoted as Search Enhanced Summarization (SES). In the SES technique, the bug-revealing log message, commonly referred to as the summary, is first extracted from the fault’s stack trace. Then, the predicted summary is used to query a pre-populated Retrieval Augmented Generation (RAG) vector database to retrieve one, three, or five matching records. These matches to the summary query vector contain the potential source code method name responsible for generating the log message query. At least one of the retrievals approximates the fault location.

Author

• Edoigiawerie, Osazuwa

Language

• en

Keyword

• Fault Diagnosis
• Fault Localization
• Machine Learning
• Artificial Intelligence
• Software Fault
• Fault Categorization

Date created

• 2025

Type of Work

• Dissertation

License

• All rights reserved

Rights statement

• In Copyright

GW Unit

• Engineering Management

Degree

• D.Engr.

Advisor

• Alsarhan, Hamza F

Committee Member(s)

• Abreu, Kevin
• Alsarhan, Hamza F
• Akinfaderin, Adewale

Persistent URL

•  https://scholarspace.library.gwu.edu/etd/d504rm35x