Generic Approach for Interpretation of PCA Results - Use Case on Learner's Activity in Social Media Tools

Abstract

Intensive usage of social media tools for educational purposes transformed many previously tackled issues from classical e-Learning systems. Among the most general challenging issues reside in building classification models having the performed activities set as independent variables and final grade as dependent variable. A critical step in data analysis process regards building interpretable models in terms of explaining feature values and ranges along with their influence on target class. We asked whether dimensionality reduction techniques may be effectively used such that high quality interpretable models are obtained. Principal component analysis (PCA) dimensionality reduction technique, scaling and several classical classification techniques were used to create a data analysis pipeline that produces classification models with similar accuracy of initial classification models built on raw available data. Experimental results show that features that characterize the activity performed on each social tool and on all tools are highly interpretable in our classification context. The proposed approach is flexible and can be adapted to similar practical use cases in which a large number of features is difficult to be interpreted and a digest is required as being more useful for bringing a better insight on data.

Keywords

• data engineering
• knowledge representation
• machine learning
• social network services
• social computing