Mandelbug Classification Engine: Transfer Learning and NLP Approach

Classifying bugs in software systems indeed often involves considering factors like severity, complexity and reproducibility. More elusive and troublesome types of bugs in software development are Mandelbugs which exhibit characteristics of being both complex and non-deterministic, making them exceptionally challenging to reproduce and resolve. However, developers can perform a quick, inexpensive yet most effective methods to identify Mandelbug root causes, and design targeted fault-tolerance mechanisms to enhance system reliability and resilience. His work studied the distribution of Mandelbugs and proposed a classification engine – machine learning, feature engineering, transfer learning and natural language processing (NLP) approach to quickly and effectively categorize Mandelbugs. We evaluated our proposed solution by extracting and processing the text descriptions of Mandelbugs obtained from four different datasets which has 210 Mandelbug records. Our performance evaluation revealed that use of transfer-learning approach has improved F1-scores as well as accuracy (30% - 65%) when compared to that of baseline classifiers.