Using item response theory in machine learning algorithms for student response data

Item Response Theory (IRT) is widely used to solve the problem measure of the hidden capabilities of a student based on historical data. But these methods have their limitations, which do not allow to obtain accurate results. In the paper, we propose combining IRT with machine learning algorithm to improve accuracy. IRT calculates the general abilities of students and grouping of students by abilities, this information is an additional feature for the machine learning algorithm. Three methods of machine learning were compared: logistic regression, XGBoost, LightGBM. LithtGBM performed best on three recall, precision, f1 metrics. This method can be used for adaptive learning systems that will allow students to choose the correct and fastest direction in learning subjects.