Machine Learning Algorithm, Scaling Technique and the Accuracy: An Application to Educational Data

Machine learning (ML) applications in educational data mining have become an increasingly popular research area. Literature indicates a lack of research investigating the impact of data scaling techniques, ML algorithms, and the nature of data on the classification's accuracy. This study aims to fulfill the above. In that direction, we use three linear and three non-linear ML classifiers and six scaling techniques to evaluate the impact of the data scaling technique and the ML algorithm on four selected educational datasets. According to the experimental outcomes for data set #1, classification accuracy was significantly influenced (p-value < 0.01) by the nature of the data. All the performance indicators except detection rate and prevalence were highly influenced by the type of ML technique used for the classification. Furthermore, there was a significant (p-value < 0.05) interaction impact of two-way interactions of the nature of the data and the type of ML technique for F1 value and balanced accuracy. Further analysis indicates that the classification accuracy varies with the level of the class variable.

I. Introduction

The development of technology, the popularity of learning management systems, and fast-growing online education generate a plethora of educational data. This makes the educational researcher's life easy due to the access to various forms of educational data. Educational data analysis has benefited from data science, enabling us to explore the relationships among the data that were not possible with the existing conventional statistical procedures.