Game-based simulation systems are increasingly being used to train users in several applications across government, in-dustry, and academia. Designing game-based training systems that can measurably improve learning while providing an engaging training experience is a challenging problem. In this paper, we describe a novel framework that tightly integrates game-based training systems with instructional com-ponents using data analysis to address this problem.Intelligent training systems based on this framework dynamically adapt both the training and the instructional components to measurably improve learning in play sessions. We propose a three phase approach to automatically identify points in a play session to predict high-value future scenarios, validate predictions, and prescribe actions. A case study using the KDD Cup 2010 educational data set is described illustrating the effectiveness of the proposed approach.