Ensemble Machine Learning One-Versus-Rest Multilevel Grade Classification and Prediction

This study investigates ensemble machine learning techniques for predicting and classifying student grades across various academic levels, including high school and bachelor’s degree predictions. It assesses nine individual models’ accuracy scores compared to ensemble models using one-versus-rest classifiers. The research aims to select the most accurate model for multilevel modern letter grade prediction. A comprehensive data preparation pipeline is implemented, and feature significance is evaluated using ordinary least regression. Nine machine learning models are employed, and their outputs are analyzed using balanced accuracy, macro-weighted accuracy scores, and execution time. The findings reveal varying levels of accuracy among the models. After using ensemble with the one-versus-rest algorithms Random Forest (84%), XGBoosting (81%), Support Vector Machine (81%), AdaBoosting (81%), Gradient Boosting (83%), Nearest Neighbors (77%), Decision Tree (67%), Logistic Regression (82%), and Bagging (83%), they exhibit different performance levels. Similarly, overall internal consistency among multilevel grade classification accuracy shows that Support Vector Machine (62%), Random Forest (65%), and Logistic Regression (60%) demonstrate the highest scores, outperforming Decision Tree (40%) and k-nearest neighbors (51%). The study suggests a one-versus-one classification approach improves balanced accuracy scores, emphasizing ensemble models’ effectiveness in accurate predictions.