The problem of accurately predicting student responses using prerequisite or diagnostic data has been a major focus of the educational research community in recent years. With practical applications in educational material and standardised test design, an enormous body of work dealing with novel strategies that provide solutions to this problem has accumulated over the years. In particular, the two most common psychometric schools of thought on this type of problem are Classical Test Theory (CTT) and Item Response Theory (IRT). While classical methods have relied more on the exploratory analysis of student data, methods inspired by IRT rely more on the probabilistic viewpoint and have gained popularity due to their shared traits with novel machine learning approaches and their historical reliability in predicting outcomes.
In this project we present a simple yet effective extension to the standard IRT modelling approach, called IRT++, which combines both the 1-parameter and 2-parameter IRT models and modulates parameter optimisation thorough simple machine learning techniques like adaptive gradient descent and random-normal initialisation of parameters.