Fast geometrical extraction of nearest neighbors from multi-dimensional data. (April 2023)

• Record Type: Journal Article
• Authors: Aziz, Yasir; Memon, Kashif Hussain
• Is Part Of: Pattern recognition. Volume 136(2023)
• Publication Date: 2023-04
• Subjects: Nearest neighbors -- Classification -- Hashing -- Windowing operation; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2022.109183
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: Nearest neighbor algorithms are the need of modern era of scientific research. Neighbourhood extraction is a lazy process due to various distance calculations. Unlike available KNN variants, proposed FGENN doesn't involve distance calculations. K parameter concept is eliminated in FGENN making it parameter free algorithm. Using window concept, FGENN geometrically extracts neighbours from data of any size. The FGENN identifies more nearest neighbours in significantly less time. Abstract: K-Nearest Neighbor (KNN) algorithm plays a significant role in various fields of data science and machine learning. Most variants of the KNN algorithm involve distance computations and a parameter (K) that represents the required number of neighbors. The recent research regarding distance computations and finding the optimal value of K have made neighborhood extraction a slow process. This research presents a fast geometrical approach for neighborhood extraction from multi-dimensional data. Instead of distance computations, the proposed algorithm creates a geometrical shape based on the number of features of data. This geometrical shape encompasses the reference data point and the neighboring points. The proposed algorithm's efficiency of time, classification, and hashing are evaluated and compared with existing state-of-the-art algorithms.