Classification rule discovery for software bug severity using knn with different distance metric

Abstract

KNN is one of the most popular machine learning algorithms. There are different parameters on which the performance of an algorithm depends. The performance of KNN algorithm depends on value of K, and distance metric. So while implementing the KNN algorithm for the classification rule discovery, a proper care must be taken to choose the value of ‘K’. Second parameter which is also very important for the KNN algorithm is the distance metric. As there are different types of the distance function which can be used with KNN, selection of distance function is also an effective task. In this paper value of ‘K’ is selected on the basis of mean error and it is found that best values of ‘K’ are 8, 12 and 15. Then using K=8, 12 and 15, KNN algorithm is implemented for bug severity classification. A comparative analysis of three distance metrics Euclidean, Manhattan and Hamming distance functions is done.