An ACT-R Predictive Model of Performance

The ACT-R computational modeling architecture has demonstrated the ability to model both recency and frequency effects in memory with much success (e.g. Anderon & Lebiere, 1998); and through the incorporation of new decay parameters at each data point, has also been shown to capture the spacing effect (Pavlik & Anderson, 2003, 2005). Stemming from the aforementioned literature, the current research sought to build an equation capable of handling the prediction of performance at later, distributed points in time, thereby breaking from the tradition of post-fitting data. As such, we integrated a single activation-based decay rate into the ACT-R General Performance Equation (Anderson & Schunn, 2000), and scaled predictions by amount of training history improvement. We tested this algorithm by extrapolating learner knowledge states from initial points in data, and predicting performance at later points in time, across different intervals of time. Implications are discussed.