Going Beyond Better Data Prediction to Create Explanatory Models of Educational Data

Abstract

PG 69 Across the vast majority of educational data mining research, models are evaluated based on their predictive accuracy. Most often, this takes the form of assessing the model's ability to correctly predict successes and failures in a set of student response outcomes. Much less commonly, models may be validated based on their ability to predict post-test outcomes (e.g., Corbett & Anderson, 1995) or pre-test/post-test gains (e.g., Liu & Koedinger, 2015). While predictive modelling has much to recommend-structs that are causally related to outcomes (Shmueli, focus is on why only that an interpretation of the data that has implications for theory, practice, or both. Here, we review educational models and, in turn, can lead to improvements to learning outcomes and/or learning theory. Educational data mining research has largely focused on developing two types of models: the statistical model and the cognitive model. Statistical models drive the formance as they learn. Cognitive models are representations of the knowledge space (facts, concepts, skills, et cetera) underlying a particular educational domain. The majority of the research reviewed here models outside the realm of cognitive model discovery that educational data mining research has produced. In the statistical modelling of educational data, approaches vary depending on whether combination of features that best predict outcomes; they are typically assessed by their causal relationships between constructs that can be either observed or inferred from the data. The vast majority of educational data mining research has focused on achieving pre-such as having parameters that map to interpretable constructs, having fewer parameters overall, and involving human input early in the model development process.