Application of Particle Swarm Optimization for Software Defect Prediction Using Object Oriented Metrics
Ruchika Malhotra, Nishant Nishant, Spandun Gurha, Vishal Rathi · 2021
With the growing number of software applications being developed for every small challenge, the importance of devising efficient software defect prediction models is imperative. Over the years, various machine learning techniques have been utilized to develop defect prediction model and have managed to achieve good results. In all defect prediction models, the task of correcting imbalanced data and feature selection has been of great significance. In this paper we have tried to analyze the working of the oversampling technique SMOTE along with feature selection using Particle Swarm Optimization on Object Oriented metrics. The selected features were then used to train the datasets one of the most popular classification techniques-Support Vector Machine to predict defects. The four datasets used for this study are of different Apache applications whose source code was obtained from open-source platforms and the raw data was pre-processed to obtain Object Oriented metrics. The performance measures used to record the results were Area Under ROC Curve, Recall and F-Measure values which showed that the Support Vector Classifier performed better on the dataset that had been balanced using SMOTE and acted upon by Particle Swarm Optimization for selecting feature set.