Using recurrent neural networks to build a stopping algorithm for an adaptive assessment

Abstract

ALEKS (“Assessment and LEarning in Knowledge Spaces”) is an adaptive learning and assessment system based on knowledge space theory. In this work, our goal is to improve the overall efficiency of the ALEKS assessment by developing an algorithm that can accurately predict when the assessment should be stopped. Using data from more than 1.4 million assessments, we first build recurrent neural network classifiers that attempt to predict the final result of each assessment. We then use these classifiers to develop our stopping algorithm, with the test results indicating that the length of the assessment can potentially be reduced by a large amount while maintaining a high level of accuracy.