The Comparison of Linear Regression Method and K-Nearest Neighbors in Scholarship Recipient

Abstract

The scholarships award for students are often subjective, not transparent, un-measurable, and less precise on target. One of the computer technologies used to process big data such as scholarship recipient problems is data mining. Various methods of data mining can be used to predict the feasibility of data such as K-Nearest Neighbors (KNN) and Linear Regression. This study compares both methods in solving the scholarship recipient problem. The attributes used are Semester period, Grade Point Average (GPA), Statement Letter of Active Student, Letters of Assistance, Student Identity Card, Identity Card, Family Card, Study Result Card, Statement Letter, Bank Account, and Statement Letter of Passed Administration. The variables used in the comparison process include Accuracy, Precision, Recall, Classification Error, Absolute Error, and Root Mean Square Error (RMSE). Data from 8212 scholarship recipients are tested through simulation testing of training data and testing data 90:10, 70:30, 50:50, 30:70, and 10:90. Herein, Rapidminer is used as a tool to view the results of analysis from both methods. As a result, both methods for data simulation 90:10 and 70:30 provide 100% of accuracy, precision, and recall. Meanwhile, KNN from data simulation 50:50, 30:70, and 10:90 provide better performance in accuracy, precision, recall, classification error, absolute error and RMSE than in Linear Regression with comparison of mean differences are 17.79%, 18.1%, 10.83%, 17.79%, and 0.25 respectively. KNN and Linear Regression methods have been successfully applied to classify and cluster the data of scholarship recipients. The result has shown that KNN method is more effective and efficient rather than Linear Regression method. This provides new knowledge contribution. Hopefully, the selection process of scholarship recipients can be implemented much better, transparent, no longer subjective, and right on the target.