A model for student action prediction in 3D virtual environments for procedural training

Abstract

This paper presents a predictive student actionmodel, which uses student logs generated by a 3D virtual environment for procedural training to elaborate summarized information. This model can predict the most common behaviors by con- sidering the sequences of more frequent actions, which is useful to anticipate common student? errors. These logs are clustered based on the number of errors made by each stu- dent and the total time that each student spent to complete the entire practice. Next, for each cluster an extended au- tomata is created, which allows us to generate predictions more reliable to each student type. In turn, the action pre- diction based on this model helps an intelligent tutoring sys- tem to generate students? feedback proactively.