A Study of Hyper-Parameter Tuning in The Field of Software Analytics

摘要

Software defect is defined as the divergence from the expected or anticipated behavior of the software. Humans can manually detect the defects but this process is always time-consuming and tiresome. To save human effort, machine learning algorithms are used to predict faulty modules in our software. One of the major aspects which are overlooked is the performance measure of these algorithms. Defect Predictors can be optimized by tuning the parameters of the algorithm. The defect predictors are easy to use. The paper attempts to analyze and compare various methodologies to tune the defect predictors. The research papers which are analyzed here have used data-set from the PROMISE repository, open-source JAVA systems, SEACRAFT repository to train and test our methodologies. It is concluded that tuning the defect predictor is not a complicated task, and it enhanced the performance of the defect predictor to a great extent. Thus it becomes imperative to tune the defect predictors using optimization algorithms before using them for detecting anomalies in our software.