Hybrid Differential Evolution and Tabu Search for Parameter Tuning in Software Defect

Software Defect Prediction (SDP) helps in identifying defective modules as early as possible in the Software Development Life Cycle (SDLC). Machine learning (ML) methods are employed as defect predictors to provide insights into a software’s health. However, without fine-tuning the parameters of these defect predictors, the insights can be inaccurate. Hence it becomes crucial to tune defect predictors and optimize performance. The goal of this work is to discuss and compare some basic and easy strategies for parameter tuning. To work with Tabu Search (TS), Differential Evolution (DE) and Hybrid Differential Evolution and Tabu Search (HDETS) as optimizers, we used datasets from open-source JAVA systems. The parameter tunings were then applied to the test datasets to compare and contrast the outcomes of three optimization strategies. We discovered that tuning increases performance in a large number of cases. Since there was a considerable improvement in performance, after parameter tuning, we can say that when predicting defects in software, it is insufficient to just present the outcome without conducting a thorough tuning optimization analysis. Additionally, it was revealed that DE, TS and HDETS did not produce the same outcomes. As a result, for the vast number of datasets, it becomes essential to conduct tuning using a variety of optimization strategies to arrive at the best potential outcomes.