Researches On Software Defect Prediction Model Based On Attribute Discretization

Entering the information age, almost no one from all walks of life can work without using computer software. Nowadays, the software products can not meet people’s demand of high quality software, how to design and develop high quality software products effectively becomes an attractive research area. Software defect prediction is an effective tool for developing high quality software, data mining algorithms were used for constructing software defect prediction model on historical data which can help developers to predict whether a software module contains defect, high-quality software systems can be developed with limited resources by this way. Now software defect prediction is one of the most active fields of software engineering, and it plays an important role in the analysis of software quality and for improving the efficiency of software development. In this paper, a software defect prediction model based on the Naive Bayes was designed to improve the accuracy of prediction. In addition, an effective discretization strategy according to the attribute discretization was also used.First, this paper provides an overview of software defect prediction technology’s research background, significance and current situation. Data mining algorithm for software defect prediction and the attribute discretization and common discretization methods were also introduced and analyzed in detail.Then, this paper gives an introduction about Naive Bayesian model which has the advantages of simple, high precision, high speed and so on. The combining algorithm of Particle swarm optimization and Naive Bayesian classification inspired by that model for constructing software defect prediction model was proposed in the next.Finally, considering entropy principle approach can not quickly find the best separated point, the Particle Swarm Optimization algorithm and expected entropy which discrete the attributes by expected entropy guiding particle swarm optimization to search the best segments was proposed, and it can intelligently and quickly partition the attributes. Based on feature selection, software defect prediction model on original datasets and discrete datasets was built respectively, the simulation experiments show that the proposed method of discretization is effective and feature selection technique is necessary.