Weighted Residual Target Proximity Kernel Pursuit Regression based Students Admission Prediction for Higher Education

Abstract

 Education plays a significant role in providing individuals with the knowledge, skills, and tools needed for personal as well as academic growth. Due to the increasing number of higher education graduates, student admissions process is essential for selecting qualified candidates for admission in a universities or colleges. An admissions system with suitable and reliable criteria is important to select students who performing well academically as well as other activities at institutions. Therefore, each university or college needs to use the best possible techniques for analyzing the history of a student's academic performance and other extracurricular activities before admitting them. Education Data Mining (EDM) involves the application of data mining techniques to large educational databases with the aim of discovering useful information.  Several machine learning techniques have been developed in this area, but there are issues related to time efficiency and errors in prediction of admissibility for higher education. To address the aforementioned challenge, a novel technique named Weighted Residual Target Proximity Kernel Pursuit Regression (WRTPKPR) has been developed. This technique aims for the accurate prediction of graduate admissions with minimal error by mapping the course to the students based on their interests and CGPA secured. The proposed WRTPKPR technique includes three major phases namely data acquisition, preprocessing, and feature selection for accurate predictive analytics.Top of Form  The WRTPKPR technique initiates by collecting information from the dataset during the data acquisition phase.  Following data acquisition, the WRTPKPR technique undergoes data preprocessing to transform the input data into a suitable format for accurately predicting whether the student is admissible or not.  Two key processes are conducted in the data preprocessing phase, namely, missing data imputation and outlier data detection. In the initial step, the Horvitz–Thompson Weighted imputation method is applied to generate missing data points based on other known data points in the dataset. In the second step, an outlier detection method based on the maximum normalized residual test is employed to identify data points that significantly deviate from the rest of the data point in the dataset.  With the preprocessed dataset, the target feature selection process is conducted by applying Kernel Cook's Proximity Projection Pursuit Regression. Based on the selected target features, accurate admission predictions are made for higher education graduates with minimal time consumption. Experimental evaluation considers factors such as admission prediction accuracy, precision, recall, F1-score and admission prediction time. The results demonstrate that the proposed WRTPKPR technique achieves efficient performance outcomes, including higher accuracy, precision, with minimized time.