DefinePK

Abstract

Software defect prediction is very essential in the field of software development and maintenance which is highly demanded quality of service. Heterogeneous defect prediction is the most appropriate method for real time datasets. The heterogeneous metrics of cross projects are used for predicting in many existing models, but the presence of outlier and noisy datasets are not considered as an important factor thus the standard prediction models face challenges in producing more accurate results. This paper focuses on handling the impreciseness and vagueness in treating noisy and outliers in software defect prediction dataset. This is accomplished by adapting bipartite ranking based feature ranking, which converts the target attribute size same as source attribute size and the feature selection by selecting the top attributes. The noisy and outlier is handled by kernel neutrosophic clustering by introducing the degree of truthiness, indeterminacy and falsity. Finally, Grey Wolf Optimization enhances the heterogeneous cross project prediction process by selecting the significant centroids in kernel neutrosophic clustering unlabeled instances. This work used six different heterogeneous datasets for software defect prediction and the results explores that the proposed model performs better and increase the prediction rate prominently.