Accurate software maintenance effort prediction is a very important task for the software development life cycle. Ensemble effort predictions aggregate the outputs of various weak learners and form a powerful model. In this paper, the authors propose a reliable software maintenance effort prediction model using a heterogeneous ensemble model. This model is evaluated on two publicly available datasets. The popular homogeneous models, random forest, bagging, random subspace, AdaBoost and light boost, are used to form a heterogeneous ensemble model. The proposed heterogeneous model outperforms all the base models by showing the lowest value of error (measured in terms of MdMRE) as well as the highest value of accuracy (measured in terms of Pred(.25)). Heterogeneous models built from the various homogeneous models will work effectively for predicting maintenance efforts.