Application of Two-Stage Data Pre-processing Approach for Software Fault Prediction

Software fault prediction is a valuable exercise in software quality assurance to better allocate limited testing resources. Classification is one of the effective strategies for predicting software errors. The classification models are trained based on data sets obtained by historical repositories of mining software. In this project, a new Two-stage data preprocessing approach is applied with classification models such as Naive Bayes, Decision Tree, Knn Classifier and SVM to improve the prediction accuracy of each classification model. The data preprocessing approach in two stages incorporates both the selection of features and the reduction of instances. Specifically, in the feature selection stage, first relevance analysis is done, second, a threshold-based clustering method is proposed, termed novel threshold-based clustering algorithm, to drive redundancy control. In the instance reduction stage, random sampling is applied to maintain the balance between defective and non defective instances. To demonstrate this project chose real-world software project dataset, such as Eclipse.