Intelligent systems are getting more popular for predicting student success and obtaining useful insights from educational data. Traditional educational institutions fail to fully utilize their student's data to enhance their instructional setting. Interpreting large amounts of educational data may be quite essential for institutions. For instance, such intelligent systems are needed in order to address questions like "how many credits should a particular student be allowed in a semester in order to improve his/her Grade Point Average (GPA) with no or minimal extension on his/her graduation time?" or “what students in a class should be proactively offered additional tutoring to increase the success rate of a class?”. A fundamental problem that needs to be dealt with while addressing such questions is predicting a student’s success in a particular course ahead of time. In this paper, we propose a general clustering-based framework to estimate student course grades before the course commences. The clustering-based framework allows the grouping of similar students/courses in a dataset prior to fitting a learning model. Using a separate model for a group of similar samples can significantly increase their accuracy levels, as it is much easier to find patterns from a set of similar items. We demonstrate the application of the proposed grade prediction model on a real-life dataset. The experimental results show that our proposed framework outperforms the state of the art by approximately 36% improvement in error rates.