Research Information System
10th International Conference on Computational and Experimental Science and Engineering, Antalya, Turkey, 27 - 30 October 2023, pp.202, (Summary Text)
The k-Nearest Neighbors (kNN) algorithm is a fundamental tool in machine learning, yet its
performance hinges on the accurate computation of distances, often via the Euclidean
distance metric. This research seeks to enhance the kNN algorithm by optimizing this metric
through an adaptive model that adjusts the distance calculation based on dataset-specific
attributes, improving upon the rigidity of the traditional Euclidean metric.
Our method, when applied to a unique sovereign credit rating dataset, showed significant
improvements in predictive accuracy and computational efficiency, validating the efficacy of
this approach. Our model extends beyond traditional kNN by integrating feature importance
weighting from the Random Forest algorithm, which provides a nuanced measure of sample
proximity and improves classification performance.
The results of this research have far-reaching implications for machine learning and data
mining, particularly in high-stakes prediction accuracy scenarios. They highlight the
potential of dynamic distance metrics and hybrid models in enhancing algorithmic
performance. Future research will focus on validating the effectiveness of this optimized
distance metric across diverse datasets, affirming the versatility of our adaptive model and
its implications for the kNN algorithm.