Modeling the development of problem solving skills in chemistry with a Web-based tutor

This research describes a probabilistic approach for developing predictive models of how students learn problem-solving skills in general qualitative chemistry. The goal is to use these models to apply active, real-time interventions when the learning appears less than optimal. We first use self-organizing artificial neural networks to identify the most common student strategies on the online tasks, and then apply Hidden Markov Modeling to sequences of these strategies to model learning trajectories. We have found that: strategic learning trajectories, which are consistent with theories of competence development, can be modeled with a stochastic state transition paradigm; trajectories differ across gender, collaborative groups and student ability; and, these models can be used to accurately (>80%) predict future performances. While we modeled this approach in chemistry, it is applicable to many science domains where learning in a complex domain can be followed over time.