Using Principal Component Analysis to support students' performance prediction and data analysis

Abstract

We propose a methodology based on Principal Component Analysis (PCA) for predicting students' performance and for identifying relevant patterns concerning their characteristics. The proposed methodology starts by data preprocessing, dimensionality reduction by means of PCA, classification and the results' interpretation. Besides studying the prediction capability of the proposed methodology, we also investigate the effectiveness of PCA to interpret data patterns in educational data. The proposed methodology was validated using two public datasets describing students achievements, as well as their social and personal characteristics. Experiments were conducted by comparing the predictive performances of the high dimensional dataset with low-dimensional reduced datasets. The results showed that PCA retained relevant information of data after the dimensionality reduction and it showed to be useful for identifying patterns in students' data.