Software change prediction plays a significant role in efficient allocation of limited resources on the identified problematic classes which are prone to changes. Managers can allocate the resources on change-prone classes, thus preventing the changes or defects to propagate to the later phases or the next release of software. This reduces the maintenance effort resulting in delivery of a better quality software. Ensemble models which aggregate the weak machine learning models to improve the performance have shown promising results in the area of software change prediction. However, due to the limited work on the application of heterogenous ensembles to predict change-prone classes, this study explores four such ensembles (Average Probability Voting, Majority Voting, Stacking and Grading) which have been aggregated using three weak non-ensemble classifiers. The performance of the models has been compared with the constituent base classifiers. The empirical validation has been done using seven open-source software datasets. The results suggest the use of heterogenous ensemble with most of the datasets giving G-mean value greater than 51.22 and balance value greater than 54.67.