The efficiency and accuracy of defect control are critical components in software testing, as they determine the final product's quality and cost. Rejection of defects for various reasons, like non-reproducibility, erroneous classification or inadequate information, is one of the largest issues that testers face. This paper presents an AI-driven approach that reduces the number of defect rejections by using the past defect data to give testers real-time advises and warnings.
When a tester reports an issue, the model looks at the problem's description and title, making inferences and recommendations based on historical data to increase the fault's correctness. This feedback strategy reduces rejection rates and increases the overall efficiency of defect management by helping testers resolve potential issues before submitting a defect.
The recommended solution involves training an AI model on a large dataset of previous defects, which includes details on DefectTitle, Description, ResolutionCategory, ReasonForRejection (optionally with RejectionCategory) and ResolutionDescription to be more accurate in its recommendation. The model uses natural language processing (NLP) technique to transform textual data into meaningful numerical representations. BERT (Bidirectional Encoder Representation for Transformers) language model is deployed for understanding the context and semantics of defect titles and descriptions. Next, a binary classification technique- Random Forest Algorithm is used to predict if the issue could be accepted or denied. Regular retraining and continuous monitoring ensure that the model remains accurate and flexible over time.
This paper covers in depth the deployment process, model training, data preparation, and real-world implications of deploying this system in a real software testing environment. This strategy solves a common software testing problem in a scalable and efficient manner by utilizing artificial intelligence.