Finding key integer values in many features for learners' academic performance prediction

In recent years, along with the proliferation of the learning management system (LMS), a large amount of data regarding the interaction between the system and the learners has been accumulated. Correspondingly, various data mining methods have been applied to these data. In order to employ a suitable computational model that is at the core of the data mining method and is not automatically acquired by the mining method itself, it is important to make or find various reasonable hypotheses for target variables. In this paper, we propose a method for analyzing closely the degree to which the explanatory variables represented in integer value contributes to predicting categorical objective variables, such as a learner's academic performance. Specifically, we describe that a decision tree combining support vector machines (SVM) achieves accuracy consistent with existing research, and it contributes further extraction of particular explanatory values from the integer features. Before making a model with SVM, our proposal method expands original features represented by integer value to corresponding binary features. With this expansion of original features, we can identify the key values that closely relate to a learner's academic performance from behavioral features gathered from LMS. Identifying such key values in specific features plays an important role in developing a hypothesis that explains the objective variables, using them as explanatory variables. We believe that closer analysis of these key explanatory values will find latent knowledge that can improve learners' academic abilities.