Can We Predict the Correctness of Problem-Solving Question Using Process Data

• In Event: Applications of Artificial Intelligence in Education

Abstract

Problem-solving tasks are developed for testing students' skills of processing information, extracting existing knowledge, and finding a good solution. Computer-based problem-solving tasks provide opportunities to look at students' behaviors during tasks. The present study extracts potential features from process data and uses supervised machine learning techniques to predict the correctness of the final answer. Naïve Bayes, Support Vector Machine, Decision Tree, and Random Forest are conducted to train the models. The results show that the behaviors during the task could predict students' final choices. Random forest performs best in cross-validation. However, when applying the models on the testing dataset, support vector machine classifier performs best among these four algorithms. The results also indicate some inflation in cross-validation.

Author

• Bixi Zhang, The City University of New York