Performance Analysis of Datamining Algorithms for Software Quality Prediction

Data mining techniques are applied in building software fault prediction models for improving the software quality. Early identification of high-risk modules can assist in quality enhancement efforts to modules that are likely to have a high number of faults. Classification tree models are simple and effective as software quality prediction models, and timely predictions of defects from such models can be used to achieve high software reliability. In this paper, the performance of five data mining classifier algorithms named J48, CART, Random Forest, BFTree and Naive Bayesian classifier (NBC) are evaluated based on 10 fold cross validation test. Experimental results using KC2 NASA software metrics dataset demonstrates that decision trees are much useful for fault predictions and based on rules generated only some measurement attributes in the given set of the metrics play an important role in establishing final rules and for improving the software quality by giving correct predictions. Thus we can suggest that these attributes are sufficient for future classification process. To evaluate the performance of the above algorithms Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Receiver Operating Characteristic (ROC) and Accuracy measures are applied.