Learning Analytics through a Topological Data Analysis Approach.

Inoue, H., Yasutake, K., Yamakawa, O., Tagawa, T., Sumiya, T. & Okamoto, K.

EdMedia + Innovate Learning

Association for the Advancement of Computing in Education (AACE)

Vienna, Austria

Target data for learning analysis covers a wide range of data, including learning status recorded in the learning management system, scores for quizzes and assignments, final grades, and behavior in the network. When the data is large and complex, it may be challenging to analyze and extract essential information. For example, it is only realistic to display data in up to three dimensions in visualization. In the case of principal component analysis, multidimensional data must be reduced to lower dimensions. This study views the data as a topology to solve this problem That is, we apply topological data analysis, which treats the shape of the data algebraically. The reason is that if the data is quantified and proximity between information can be defined abstractly, analysis is possible within the same framework. This study attempts to apply topological data analysis as a new method of learning analysis. We analyzed persistent homology in an open educational dataset and examined the characteristics of data classified by grade. The results suggest that the high and low grade groups have the same topological structure, whereas the middle-level group has a different topological structure.