Developing Predictive Models for Computational Thinking Skills of Lower Secondary School Students using Machine Learning Techniques

Abstract

Learning outcomes are classified as big data and there is not much research on predicting students' performance at lower secondary school levels. The objectives of this research were to develop predictive models for computational thinking skills of students using machine learning techniques, and to compare the effectiveness of the predictive models for computational thinking skills. The predictive model development utilized a training dataset of 143 records from the basic subject grades of lower secondary students in a school located in Suratthani province. The independent variables consisted of grades in 10 basic subjects, while the dependent variable was the level of computational thinking skills to be predicted.  This research applied a Cross Industry Standard Process for Data Mining (CRISP-DM) framework to develop the predictive models using three machine learning techniques: Naive Bayes, Decision Tree, and K-Nearest Neighbor, and the research also examined such model effectiveness using a 10-fold cross validation technique. The results of the model development to predict computational thinking skills found that the predictive models developed from Naive Bayes, Decision Tree, and K-Nearest Neighbor techniques had an accuracy of 60.05%, 74.95%, and 68.48%, respectively. The results of comparing the model performance to predict computational thinking skills revealed that the predictive model developed from the Decision Tree technique could predict the level of computational thinking skills most effectively among the three models.