Software Change Prediction using Ensemble Learning on Object Oriented Metrics

In order to reduce developers’ time testing the defects in software, this research work has come up with an ensemble learning approach using various machine learning and deep learning classifiers. This empirical study starts with the data collection process. Own dataset has been created by using Understand tool and further it will be applied over sampling techniques. The data set has a binary class output, and thus by applying data over sampling techniques, the number of instances of both classes remain comparable. After that various classifiers are trained with the entire training data set. After this step, the proposed model will predict the outcome that these classifiers would create using the training set and compare them with the actual values of the training dependent variable. After this the information that claim which of the classifiers were able to correctly predict the outcome will be given a data point. After that the neural networks will be trained for each of the classifiers with input variables as training independent variables and output variables as the previously recorded data as to whether a classifier correctly classifies a data point or not. This predicting network would tell us whether the classifier corresponding to that particular network would correctly classify a particular data point or not. To predict the outcome of any new data point, it will be fed to the predicting network and find which classifier would correctly predict the outcome and the aggregated result of the selected classifiers would be reported.