Improving Predictive Accuracy of Models of Learning and Retention Through Bayesian Hierarchical Modeling: An Exploration with the Predictive Performance Equation
- Michael Collins, ORISE, Air Force Research Laboratory, Dayton, Ohio, United States
- Florian Sense, Infinite Tactics, Dayton, Ohio, United States
- Michael Krusmark, L-3 Communications, Wright Patterson Air Force Base, Ohio, United States
- Tiffany Jastrzembski, Air Force Research Laboratory, Wright-Patterson Air Force Base, Dayton, Ohio, United States
AbstractHuman learning has been characterized by three robust effects (i.e. power law of learning, power law of decay, and spacing), which have been empirically validated across multiple domains and time intervals. To account for these different effects mathematical model of learning and retention have been developed. These models hold a great deal of potential for application a wide range of educational and training scenarios. However, many models are not validated according for their ability to make accurate predictions of human performance. The predictive ability of these models is made increasingly complex by the needs of training domain, needing both to predict both skill decay and reacquisition from little historical data. In this paper, we examine the predictive capability of the Predictive Performance Equation (PPE) implemented in a Bayesian hierarchical model. Through a comparison of two Bayesian hierarchical models we show how hierarchical model fit to a participant’s performance across a set of items compared to only a single item improves PPE’s predictive accuracy of both skill decay and reacquisition over multiple learning schedules