Comparative analysis of k-nearest neighbor and modified k-nearest neighbor algorithm for data classification

Abstract

Data mining is the process of handling information from a database which is invisible directly. Data mining is predicted to become a highly revolutionary branch of science over the next decade. One of data mining techniques is classification. The most popular classification technique is K-Nearest Neighbor (KNN). But there is also the Modified K-Nearest Neighbor (MKNN) classification algorithm which is the derived algorithm of KNN. In this paper we will analyze the comparison of KNN and MKNN algorithms to classify the data of Conditional Cash Transfer Implementation Unit (Unit Pelaksana Program Keluarga Harapan) which consist of 7395 records. Comparative analysis is based on the accuracy of both algorithms. Before classification, K-Fold Cross Validation was done to search for the optimal data modeling resulted in data modeling on cross 2 with accuracy of 93.945%. The results of K-Fold Cross Validation modeling will be the model for training data samples and testing data to test KNN and MKNN for classification. Classification result produced accuracy based on the rules of confusion matrix. The test resulted in the highest accuracy of KKN by 94.95% with average accuracy during the test was 93.94% and the highest accuracy of MKNN was 99.51% with the average accuracy during the test was 99.20%, almost all testing from the first test up to the tenth, MKNN algorithm is superior and has better accuracy value than KNN so it can be analyzed that the ability of MKNN algorithm in accuracy is better than KNN. It can be concluded that MKNN algorithm is capable of handling accuracy better for classification than KNN algorithm, by ignoring other aspects such as computerization, time efficiency, and algorithm effectiveness.