The development of high-quality software at lower cost has always been the main concern of the developers as well as of the users. Eliminating the defects in software at the initial development stage can increase quality and reduce the overall cost. Testing only those modules which are likely to be defective are helpful for development team to manage and use resources effectively. Many machine learning-based frameworks have been proposed for the prediction of software defects in initial development stage however accuracy evaluation of proposed techniques on benchmark datasets was lacked. In this research, we proposed a framework for the prediction of software defects using ensemble learning and feature selection techniques by using WEKA. The accuracy of the proposed model has been evaluated by using publicly available cleaned NASA datasets. Moreover, the results have been compared with the widely used advanced classification techniques. The Proposed framework consists of five stages. First stage is dealing with the extraction of relevant dataset. Second stage is dealing with variants of base classifiers and selection. The base classifiers include: ?Decision Tree (DT), K-nearest neighbor (kNN), Naive Bayes (NB), Random forest (RF) and Support Vector Machine (SVM)?. Pre-processing and feature selection have been done in third stage. In fourth stage, we used stacking technique to create an ensemble of the classifier-variants, which have performed well in third stage. Fifth stage deals with the results and performance evaluation by using different measures including: ?Precision, Recall, F-measure, Accuracy, MCC and ROC?.