Prediction Intervals for Performance Prediction

The Predictive Performance Equation (PPE) is a mathematical model of learning and forgetting developed to capture performance effectiveness across training histories, and to generate precise, quantitative point predictions of performance by extrapolating the unique mathematical regularities indicative of the learner. This equation is implemented in the Predictive Performance Optimizer (PPO) cognitive tool, designed to help learners and instructors make principled training decisions through examination of the learning and retention tradespace. Because the point predictive nature of the model implies a high degree of certainty, decision-makers could be misled into making less than optimal decisions in applied settings; and with regards to basic science, the model lacks prediction error and uncertainty which would more accurately represent the predicted range of human performance. Implementation of prediction intervals into a point predictive model of human performance is unprecedented in the psychological literature. We must balance the competing factors of reduced performance variation as practice accumulates, and greater prediction uncertainty as time spans increase. In this paper, we explore new methodologies for incorporating prediction intervals into quantitative predictions of future performance.