Prediction of Undergraduate Student’s Study Completion Status Using MissForest Imputation in Random Forest and XGBoost Models

Abstract

The number of higher education graduates in Indonesia is calculated based on their completion status. However, many undergraduate students have reached the maximum length of study, but their completion status is unknown. This condition becomes a problem in calculating the actual number of graduates as it is used as an indicator of higher education evaluation and other policy references. Therefore, the unknown completion status of the students who have reached the maximum length of study must be predicted. The research compared the performance of Random Forest and Extreme Gradient Boosting (XGBoost) classification models in predicting the unknown completion status. The research used a dataset containing 13.377 undergraduate students’ profiles from the Higher Education Database (PDDikti), Ministry of Education, Culture, Research, and Technology. The dataset was incomplete, and the proportion of missing data was 20,9% of the total data. Because missing data might lead to prediction bias, the research also used MissForest imputation to overcome the missing data in the classification modelling and compared it to Mean/Mode and Median/Mode imputation. The results show that MissForest outperforms the other two imputations in both classifiers but requires the longest computation time. Furthermore, the XGBoost model with MissForest is significantly superior to the Random Forest model with MissForest. Hence, the best model chosen to predict the completion status is XGBoost with MissForest imputation.

Prediction of Academic Performance of Engineering Students by Using Data Mining Techniques

“…The present study may conclude that the proposed model achieved the highest accuracy, precision, and recall for predicting poor-performing students. The present study shows that the ensemble model performs better than the single base classifier and is consistent with the findings of previous researchers [28]- [30].…”

An Investigation into the Variables Affecting School Principals Burnout through XGBoost Method

“…Woo and Kim (2022) investigated the effect of learning orientation on the modelling stage based on gender via XGBoost model. Nirmala et al (2022) compared the performance of Random Forest and XGBoost models in predicting the completion status of the students reaching the maximum length of study.…”

Association Rules in Random Forest for the Most Interpretable Model

“…The random Forest algorithm in this study was conducted on the default hyperparameter, that was, 𝑚𝑡𝑟𝑦 = √𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑟𝑒𝑑𝑖𝑐𝑡𝑜𝑟𝑠 = 5 and 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑡𝑟𝑒𝑒𝑠 = 100. The value of 𝑚𝑡𝑟𝑦 defined the number of predictors involved in the best splitting [1]. Table 6 shows the performance measures of the resulting random forest.…”

“…Each decision tree in a random forest is constructed independently. The final random forest prediction for the classification case is based on aggregating predictions with the majority of votes from all decision trees [1]. Random forest is one of the most popular ensemble methods because random forests can be applied to various prediction problems and produce competitive accuracy.…”