On Enhancing The Performance Of Nearest Neighbour Classifiers Using Hassanat Distance Metric

TL;DR

This paper demonstrates that the Hassanat distance metric improves the accuracy and robustness of nearest neighbor classifiers, outperforming traditional distances and being invariant to data scale, noise, and outliers.

Contribution

The study introduces and validates the Hassanat distance metric as a superior alternative for nearest neighbor classifiers, showing significant accuracy improvements and robustness.

Findings

Hassanat distance outperforms Manhattan and Euclidean distances.

Accuracy of ENN and IINC increased by over 3%.

Hassanat distance is invariant to scale, noise, and outliers.

Abstract

We showed in this work how the Hassanat distance metric enhances the performance of the nearest neighbour classifiers. The results demonstrate the superiority of this distance metric over the traditional and most-used distances, such as Manhattan distance and Euclidian distance. Moreover, we proved that the Hassanat distance metric is invariant to data scale, noise and outliers. Throughout this work, it is clearly notable that both ENN and IINC performed very well with the distance investigated, as their accuracy increased significantly by 3.3% and 3.1% respectively, with no significant advantage of the ENN over the IINC in terms of accuracy. Correspondingly, it can be noted from our results that there is no optimal algorithm that can solve all real-life problems perfectly; this is supported by the no-free-lunch theorem