Modeling NUST MS Admission Policy / Process using Machine Learning Methods

Abstract

Universities aims to identify and admit the applicants who will perform best. Therefore, they usually base their admission decisions/processes on a combination of various characteristics or measures of the applicants. Development of consensus between different universities/institutions on a common set of measures for the process of selection is not easy. However, homogeneity can be achieved by introducing dynamic features supported by empirical analysis of indigenous data. The effects of uniformity in the admission processes are manifold. It can bring harmony, respect, and better coordination between universities. Secondly, it will give confidence to the applicants with better focus to develop their attitude towards evident features of assessment. Admission policy is often seen as a quality response to upcoming generations and, consequently, higher education institutions have become a wide source of producing new talented scholars and doctors in the last decade. This research study analyzed and investigated the relationship, effectiveness and weightage distribution of different variables / parameters used in the admission process of National University of Sciences and Technology (NUST) at postgraduate level. NUST admission policy / process consists of three variables: i) interview (INT) conducted at the time of admission by a concern school/center/institution of NUST, ii) academic record of the student (ACAD), and iii) graduate assessment test (general) (GAT) (a test conducted by National Testing Service of Pakistan for higher education commission of Pakistan) or graduate record examination (general). The current weightage of INT in current admission policy is 25%, ACAD is 25% Introduction xiii and GAT is 50%. Since this research study is an empirical analysis, therefore, an archival contains the data of 13094 applicant is used. Span of the data is seven years, provided by the ICT directorate of NUST. Evidence collected from the literature review, the range and size of the data used for analysis are adequately sufficient to derive significance results and conclusions regarding the effectiveness of the process. Comparative analysis has been used to analyze the relationship of these variables between the admitted and not admitted student. Interview and academics have a statistically significant linear relationship with r =0.203, and p < 0.01. Furthermore, Cohan’s d analysis has been used to analyze the practical significance of these variables between the admitted and not admitted student. These result shows that the interview variables have a marginal difference from GAT and ACAD. Results shows that Interview (INT) is practically significant, and effect of the association is large with d = 0.704. Furthermore, principal component analysis has been used for dimension reduction and suggestion of new weightages to these variables. By using the coefficient of linear relationship between these three variables, the new suggested weightages are 36.15% for interview, 31.35% for GAT and 32.50% for ACAD. Moreover, different machine learning models are developed to check the predictive ability of these variables. Machine learning model’s includes radial based function (RBF), decision trees (DT), multi-Layer perceptron (MLP) and binary logistic regression (BLR). Results reveals that the average accuracy of these models is 63.1%. Results concluded from the predictive modeling shows that these variables are not balanced in terms of subjective weightage assigned to them. Furthermore, predictive models also concludes that these variables are not complete because of lack in predictive ability. Therefore, there is a need to review the weightages and includes other variables in the analysis like program popularity, financial stability of the applicant, location and place of residence, facility of hostel etc., Introduction xiv These all factors will provide useful insight for the suggestion of new admission policy. Further research for analyzing data of different universities can be a step towards uniform national admission policy at postgraduate level.