MACHINE LEARNING BOOSTING METHODS FOR PREDICTION A HIGHER EDUCATION INSTITUTIONS ENTRANT'S ADMISSIONS IN UKRAINE

Abstract

There is a constant and growing need for higher education institutions (HEI) to provide proper and high-quality support for the admissions campaign through information systems and technologies. Labor market trends, unreliability and low-quality sources, and a large volume of admission rules can complicate the admission process for an applicant. As a result, there is a risk that the applicant will not be able to make the right choice and quality assessment of the chances of admission. So, this paper considers increasing the entrant's chances of making an effective decision at the stage of education program selection through the implementation of an information system. The efficiency of such systems is largely based on the accuracy of its intelligent components. This article investigates the effectiveness of machine learning (ML) boosting methods to solve the problem of admission prediction through binary classification tasks. We evaluate the accuracy of such ML methods as Gradient Boosting, Adaptive Boosting (AdaBoost), and eXtreme Gradient Boosting (XGBoost). For a more detailed assessment of the studied methods, a comparison with Support Vector Machine (SVM) and Logistic regression is also presented. The simulation was performed using «Orange» software. The work of the studied methods was simulated based on a sample of archival data comprising 9,657 records of full-time entrants of two faculties of Lviv Polytechnic National University. The sample was randomly divided into training and test sets in a ratio of 80% to 20%. To ensure the reliability of the obtained result, the work of each of the studied methods was subjected to 10-fold cross-validation. Classification accuracy (AUC), Precision, Recall and F1 score performace indicators was used to analyze the results. It has been experimentally established that the highest accuracy is achieved when using XGBoost. The obtained results shows high accurate. This makes it possible to use the researched methods in the subsequent stages of building an information system to support the decision-making of applicants.