EMPIRICAL EVALUATION TO IDENTIFY THE EFFECTIVENESS OF ENSEMBLE TECHNIQUE FOR PREDICTION OF SOFTWARE FAULT

Software fault prediction is the salient factor for any software project because of the development of IT, the size of the software is increasing day by day and it becomes very difficult to test each module manually. So we need an automated approach that focus on faulty modules and enable the testing team to test the faulty modules and correct it on time. Pool of base predictors are there that are already being developed and tested. Combining the base predictors using ensemble strategy either homogeneous or heterogeneous will surely increase the prediction power of the model. Diversity of classifiers will be the primary requirement. The more diverse the classifiers are, the more efficient the model will be. So, this study demonstrates the performance of base predictors and ensemble predictors on software fault dataset and also compares them to determine which is better than the other. Various dataset from promise repository is used for comparing and implementing the model. Naïve Bayes, Logistic regression and Decision tree is used as a base predictor. Bagging, Boosting, voting, Random Forest is used as an ensemble. The results of the study prove that the prediction power of ensemble techniques is always better than the single technique.