|
University student dropout prediction improved by feature selection with Multilayer Perceptron Neural Network |
|---|---|
| รหัสดีโอไอ | |
| Creator | Nadh Ditcharoen |
| Title | University student dropout prediction improved by feature selection with Multilayer Perceptron Neural Network |
| Contributor | Anuwat Paphat, Wongkot Sriurai |
| Publisher | Faculty of Science, Ubon Ratchathani University |
| Publication Year | 2565 |
| Journal Title | Journal of Science and Science Education |
| Journal Vol. | 5 |
| Journal No. | 1 |
| Page no. | 39-48 |
| Keyword | Student dropout, prediction, feature selection, data mining |
| URL Website | https://so04.tci-thaijo.org/index.php/JSSE |
| Website title | https://so04.tci-thaijo.org/index.php/JSSE/article/view/256977/174622 |
| ISSN | ISSN 2697-410X |
| Abstract | The objectives of this research were to analyze the factors affected university student dropout prediction, to compare the efficiency of the prediction model, and to improve the model efficiency by using feature selection with multilayer perceptron neural network. Data used in this research were collected from the Office of Student Registrar and Evaluation, Division of Educational Service, Ubon Ratchathani University. They were consisted of 3 parts: (1) 1,029 records of basic data, (2) 6,826 records of semester data, and (3) 29,790 records of grade data. After data preparation process by CRISP-DM method, 882 records remained with 14 attributes. The models of university student dropout prediction were developed using decision tree, Naive Bayes, multilayer perceptron neural network, and support vector machine integrated with feature selection methods including gain ratio, Chi-square, and correlation-based feature selection (CFS). The efficiency of the generated models was measured by 10-folds cross validation to compare accuracy, precision, recall, f-measure, and mean absolute error. The experiment results revealed that the key factors to predict the university student dropout were 5 attributes including (1) grade point average (GPA), (2) GPA of courses in the student's faculty, (3) GPA of courses outside the student's faculty, (4) student loan or scholarship, and (5) student's graduation status (target class). The best model for predicting university student dropout was developed by multilayer perceptron neural network improved with CFS. The accuracy of the prediction model, after model development with parameter tuning, was 90.39%. The results indicated that the feature selection could improve the efficiency of the prediction model developed by neural network for predicting university student dropout accurately. The developed model can be further used to develop the system for predicting university student dropout. |
