Combat with Class Overlapping in Software Defect Prediction Using Neighbourhood Metric

The characteristics of data is a open problem which has been tended perceived in data analysis in machine learning research from last decades. The researcher defined some measures to identify the characteristics of the dataset by applying data complexity measures to find the fitness for purpose. The presence of class overlapping in data-sets, significantly affect performance of the classifiers. Data complexity measures provide quantitative insight in quality of the data set and overlapping existent in it. Machine learning techniques are also utilized by several researchers on healthcare datasets in software defect prediction. In this paper, our aim is to evaluates the effectiveness of new overlap measure: Near Enemy Ratio, and its effect on complexity measures and performance of the classifier. The new ration is based on nearest instances to the target instance. The experimental result offers insights in usefulness of the method and help us decide whether this solution should be applied on a particular data-set or not. © 2023, The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd.