ARTIFICIAL INTELLIGENCE AND APPLICATIONS Cluster Ensemble and Probabilistic Neural Network Modeling of Class Imbalance Learning in Software Defect Prediction

Machine learning techniques are frequent for the complicated task of predicting software defects. Often the prediction models fail to predict defects successfully as the between class imbalance increases in the training data. This paper conducts experiments on NASA KC1 dataset. Experiments were carried out on five different datasets with different class imbalance ratio, made from original KC1 dataset. Performance variations of each classifier were analyzed to detect software defects as the imbalance ratio increases for training datasets. Moreover different classifier’s performances were also investigated while trained with same datasets to find out the better one for each case. It is found that our proposed Cluster Ensemble based Probabilistic Neural Network approach has performed better than all other approaches including SMOTE based Probabilistic Neural Network for the given data sets. It is also found that incorporating Cluster Ensemble with SMOTE based Probabilistic Neural Network approach performs better than onl