Fast Duplicate Bug Reports Detector Training using Sampling for Dimension Reduction: Using Instance-based Learning for Continous Query in Real-World

Duplicate bug report detection (DBRD) is a famous problem in software triage systems like Bugzilla. It is vital to update the internal machine learning (ML) models of DBRD for real-world usage and continuous query of new bug reports. The training phase of ML algorithms is time-consumable and dependent on the training dataset volume. Instance-based learning (IbL) is an ML technique that reduces the number of samples in the training dataset to achieve fast learning for the incremental database. This research introduces a hybrid approach using clustering and straight forward sampling to improve the runtime and validation performance of DBRD. Two bug report datasets of Android and Mozilla Firefox are used to evaluate the proposed approach. The experimental evaluation shows acceptable results and improvement in both runtime and validation performance of DBRD versus the traditional approach without IbL.