||Comparative Study of Prediction Models for High School Student Performance in Mathematics
||Phauk Sokkhey;Takeo Okazaki
||Student performance; Statistical analysis technique; Machine learning algorithms; Deep belief network
||Measuring students’ performance and observing their learning behaviors are challenging tasks that can assist students and teachers in keeping track of progress in academic performance. Predicting student performance in mathematics has gained considerable attention from many researchers. Because a single tool may not be easily scalable from one context to another, several learning algorithms have been observed and compared for selecting an optimized prediction model. In this paper, we proposed a comparative study of the statistical analysis (SA) technique, machines learning (ML) algorithms, and a deep learning architecture for predicting student performance in mathematics. A proposed predictive structural equation modeling of SA, five superior classifiers in ML, and a graphical model for deep learning were executed and compared. Three datasets named, DS1, DS2, and DS3 were used in this analysis. We applied two main evaluation metrics, accuracy and predictive mean square error (PMSE), to measure the performance of the proposed models. On the three datasets, random forest produced the highest accuracy and the smallest PMSE which shows its potential as the best prediction model for the problem.