Predicting Test Scores From Gameplay Data: A Comparison of Neural Network and Regression Procedures

• In Event: Poster Session 4
  In Poster Session: Assessing Cognitive Ability and Processes

Abstract

This study examined the predictive accuracy of math test scores solely from gameplay data using either a neural network or a multiple regression procedures. Gameplay data from 56 4 and 5 year children playing 6 math games were used. The neural network procedure had an mean error of 8.3% and the multiple regression procedure a mean error of 14.7%. These results point to the promise of machine learning methods for predicting learning outcomes soley from gameplay behaviors.

Authors

• Gregory K.W.K. Chung, University of California - Los Angeles

• Charlie Parks, University of California - Los Angeles

• Katerina Schenke, University of California - Los Angeles

• Jeremy Roberts, PBS KIDS Digital