Application of classification models to predict students’ academic performance using classifiers ensemble and synthetic minority over sampling techniques

Abstract

The demand for data-driven decision making has resulted in the application of data mining in the educational sector and other disciplines. The needs for improving the performance of data mining models have been identified as an interesting area of research institutions keep a large amount of students' data, but these data are rarely used effectively in decision and or policy-making processes. This research is an attempt to enhance the performance of data mining models to predict ensemble and synthetic minority over IBK and SMO were trained and tested on 206 students' data set using previous academic performance records of Federal University Dutse, Nigeria. WEKA 3.9.1 data mining tool was used in predicting the final year student's classes of degree at an undergraduate level, while Unified Tertiary Matriculation Examination, Senior Secondary Certificate Examinations Cumulative Grade Point Average of students served as inputs to the model. The result obtained showed that on training dataset after class balancing, performing the other three (3) classifiers models in RSME (0.1098), suggesting that stacking classifiers ensemble is the best model in context of this research.

Enhancing data pipelines for forecasting student performance: integrating feature selection with cross-validation

“…In our study, we demonstrated that preprocessing filter feature selection techniques also enhanced the performance of LR in achieving comparable AUC values with GLMNET and exceeded those of RF and XGBoost. While ensemble black-box techniques have been shown to enhance the performance of DMMs (Abdulazeez & Abdulwahab, 2018;Amrieh et al, 2016;Aulck et al, 2017;Beemer et al, 2018;Lisitsyna & Oreshin, 2019;Stapel et al, 2016), several EDM studies found that non-ensemble techniques performed better than ensemble models (Adekitan & Noma-Osaghae, 2019;Bucos & Drăgulescu, 2018). A limitation to these prior studies is that they neither incorporated preprocessing filter feature selection techniques nor utilized course-specific information in their analyses; rather, they focused on demographic and student academic achievements in their data pipelines.…”

Academic success predictors for architecture students at Kano University of Science and Technology, Wudil, Kano State, Nigeria

“…Both models record non-significant Hosmer and Lemeshow test results. Compared to results from Abdulazeez and Abdulwahab (2018) as well as Aluko et al (2018) however, other EDM models are likely to produce better prediction values. We observe from our analyses that higher prediction values using BLR occur in larger proportioned samples, with lower proportions recording the highest misclassified scores.…”

“…BLR presents clear standards and statistics to check model fitness. Secondly, examples of studies utilising EDM to predict academic success within our study context test only data on previous academic records notably ordinary level results to predict undergraduate academic performance (Aluko et al, 2018;Abdulazeez & Abdulwahab, 2018). These are limitedin terms of school, socio-economic, personal, psychological and other factors influencing academic successin literature.…”