Feature Selection on K-Nearest Neighbor Algorithm Using Similarity Measure

Abstract

Data mining was the data processing technique to be obtained knowledge or important pattern of data. One of the popular methods was the KNN (K-Nearest Neighbor) which was computational simplicity. There was some weakness of KNN, vulnerable in the data high dimensionality. It was caused of data high dimensionality, so that space can be occupied in instance to be greater. The method approach would be proposed in the research was the method to be omitted several numbers of features were irrelevant to the KNN method by using similarity measures (Euclidean distance, Correlation distance and Cosine similarity). The testing was done by testing the different three datasets and computed the average of accurate results. The results of testing have successes to be omitted the data features without decreasing accuracy, that the accuracy average of feature selection using the KNN algorithm: KNN without selection was 88.804%, KNN-Euclidean distance was 89.120%, KNN-Correlation distance is 89.567% and KNN-Cosine similarity was 89.134%.