Clinical databases often consist of large number of attributes and dimensions. For clinical data analysis, some parameters are not helpful for diagnosis and prediction purposes. Therefore, applying feature selection is necessary as it removes unimportant features and improves effectiveness of the model. In this paper, a framework for disease prediction with dimensionality reduction and feature selection is proposed. The framework also uses ensemble learning methods along with some well-known classification techniques. The proposed framework is validated with five standard datasets with four different types of diseases. The comparative performance analysis shows that the proposed approach performs well with respect to some of the existing approaches. The proposed approach has the potential to be used by the practitioners for an effective prediction of other medical diseases as well.