Predicting Academic Achievement with Machine Learning Algorithms

Abstract

Education systems produce a large number of valuable data for all stakeholders. The processing of these educational data and making studies on the future of education based on the data reveal highly meaningful results. In this study, an insight was tried to be developed on the educational data collected from ninth-grade students by using data mining methods. The data contains demographic information about students and their families, studying routines, behaviours of attending learning activities, and their epistemological beliefs about science. Thus, this research aimed to solve a classification problem, two-class (successful or unsuccessful according to the exam result) was tried to be estimated from the collected data. In the study, the prediction accuracy of the supervised classification algorithms were compared and it was defined which variables were effective in the formation of classes. When the prediction accuracy of machine learning algorithms was compared, the findings indicated that the Neural Network algorithm (98.6%) had the highest score. The accuracy rate of the other algorithms are kNN (86.2%), Logistic Regression (78.4%), SVM (90.3%), Decision Tree (91.9%), Random Forest (90.0%), and Naive Bayes (81.7%). The information gain coefficient of the variables was examined to determine the factors affecting the prediction accuracy. It was revealed that demographic variables of the family, scientific epistemological beliefs of the student, study routines and attitudes towards some courses affected the classification. It can be concluded that there was a relationship between these variables and academic success. Studies on these variables will support students' academic success.

Performance Evaluation of Different Machine Learning Algorithms on Student Dataset Clustered by K-means Algorithm

“…Their study has compared various supervised classi cation algorithms like ADTree, JRip, NN, Naive Bayes, and J48, with the Neural Network algorithm achieving the highest accuracy (98.6%). [40]…”

Machine Learning-Based E-Learners’ Engagement Level Prediction Using Benchmark Datasets

“…Logistic regression is commonly employed in examining and describing the correlation that exists among entities with only two possible outcomes such as 'Yes' or 'No' and a sequence of foreseen entities [23]. Logistic Regression calculates the odds of several classes employing a boundary rationality distribution as depicted in the expression below 𝑃 = (𝑌 = 𝐾|𝑥) = 𝑒 𝑤 𝑘 * 𝑥 1 + ∑ 𝑒 𝑤 𝑘 * 𝑥 𝐾−1 𝐾…”