An improved supervised learning defect prediction model based on cat swarm algorithm

Abstract

software defect prediction has been the hot topic in the field of software engineering, for software modules that require validation and high-quality requirements, Software defect prediction provides a way to minimize nonessential software expenditures on n the premise that accurate test can be performed. On the one hand, as a kind of Swarm intelligence algorithms, the Cat Swarm algorithm is a typical Swarm intelligence algorithm that appeared in recent years, and it can combine itself with machine learning algorithms. Machine learning, on the other hand, as effective ways to set up models, ensemble learning is one of which has a better performance relative to the base learning method. In integrated learning, Random Forest based on bagging method is a method with good properties. By using this model, we can learn from a small number of software modules that are known to be defective, thus achieving the goal of predicting the defects of unknown modules, and reducing the total cost of quality assurance. In this paper, we will use cat swarm algorithm to improve the decision tree and improve the prediction effect of the entire random forest, and that using some pre-progressing methods to solve imbalanced and high-dimensional problems in datasets. From the results, we can see that the improved random forest method based on cat swarm algorithm using feature selection method has better effect than old one. In comparing our results with published benchmarks, the performance improvement can be clearly demonstrated.