Open Access
Subscription or Fee Access
Effect of Rank Based Correlation on K-Nearest Neighbour
Abstract
Abstract
Because of its simplicity, higher efficiency and effectiveness, K-nearest neighbour (KNN) isthe most popular and chosen algorithm in the last few years. It has many conveniences withthe exception of some inconvenience. In this study, we explore the correlation and dependencebetween attributes, and try to significantly improve the classification accuracy from existingalgorithm by skipping correlated attributes. To get classification accuracy, we applied thisskipping correlated attributes method in KNN ranking based classifier. In this approach, weproposed a rank minimum distance to find out the category which is unknown and take thebest decision at the result. Here used various types of real world datasets for this experiment.In implementation the results are found from original KNN algorithm and by removing someskewed and high kurtosis valued attributes. Each datasets have highly correlated attributeswhich may affect samples to classify correctly, that is why we implemented KNN algorithmwithout correlated attributes and found as expected result. The accuracy of KNN algorithmwithout correlated attributes is better than original KNN and KNN without skewed and highkurtosis valued attributes. The results indicate that without correlated attributes, the rankbased KNN method significantly outperforms than the regular k-nearest neighbor techniquefor different real-world datasets.
Keywords: Classification, correlation, rank minimum distance, K-Nearest neighbourCite this ArticleNazmul Shahadat, Abu Naem, RaselDewan. Effect of Rank BasedCorrelation on K-Nearest Neighbour.Current Trends in InformationTechnology. 2017; 7(2): 17–22p.
Because of its simplicity, higher efficiency and effectiveness, K-nearest neighbour (KNN) isthe most popular and chosen algorithm in the last few years. It has many conveniences withthe exception of some inconvenience. In this study, we explore the correlation and dependencebetween attributes, and try to significantly improve the classification accuracy from existingalgorithm by skipping correlated attributes. To get classification accuracy, we applied thisskipping correlated attributes method in KNN ranking based classifier. In this approach, weproposed a rank minimum distance to find out the category which is unknown and take thebest decision at the result. Here used various types of real world datasets for this experiment.In implementation the results are found from original KNN algorithm and by removing someskewed and high kurtosis valued attributes. Each datasets have highly correlated attributeswhich may affect samples to classify correctly, that is why we implemented KNN algorithmwithout correlated attributes and found as expected result. The accuracy of KNN algorithmwithout correlated attributes is better than original KNN and KNN without skewed and highkurtosis valued attributes. The results indicate that without correlated attributes, the rankbased KNN method significantly outperforms than the regular k-nearest neighbor techniquefor different real-world datasets.
Keywords: Classification, correlation, rank minimum distance, K-Nearest neighbourCite this ArticleNazmul Shahadat, Abu Naem, RaselDewan. Effect of Rank BasedCorrelation on K-Nearest Neighbour.Current Trends in InformationTechnology. 2017; 7(2): 17–22p.
Refbacks
- There are currently no refbacks.