A Comparative Analysis of Machine Learning Models for Prediction of Passing Bachelor Admission Test in Life-Science Faculty of a Public University in Bangladesh

The usage of neoteric technology like data mining and machine learning in the interest of the educational welfare of the students has been dramatically soared. In this work, we concentrate on the educational amelioration of university admission test candidates using state of the art data mining and machine learning methods. From the outset, a dataset has been prepared by analyzing and utilizing the limited features which are most relevant for prediction. In this regard, the data collection process is performed based on the vis-à-vis survey, and data are taken from the students who participated in the bachelor admission test exam of the Life-Science faculty of Bangabandhu Sheikh Mujibur Rahman Science and Technology University (BSMRSTU). The dataset includes 343 instances with 10 variables. As machine learning methods are robust on balanced datasets, we defined two balanced datasets generated from the original set on the basis of two mostly used re-sampling techniques. Using five selected supervised machine learning classification methods; Logistic Regression, Support Vector Machines (SVM), Artificial Neural Networks (ANN), Random Forest, and Adaboost the prediction has been performed. To evaluate our models as well as check generalization capability to previously unseen data, we specified the most significant method called Stratified K-fold cross-validation and to measure the performance mostly usable matric like AUC Score, Precision, Recall, F-Measure has been specified. By performing a comprehensive analysis, it can be included Adaboost performs best in both datasets contrariwise SVM shows second-highest performance and dataset equipped through oversampling approach become more representative.