ANALYSIS AND COMPARISON OF APPROXIMATE K-NEAREST NEIGHBOR ALGORITHMS

Abstract

This study presents a comprehensive comparison of Approximate k-Nearest Neighbor (AKNN) methods across multiple datasets, including image, text, and behavioral datasets. The performance of various AKNN algorithms is evaluated in terms of build time, search time, total time, and recall metrics for different datasets. Key findings reveal that tree-based AKNN methods exhibit vulnerability to changes in dataset contents, while graph-based algorithms demonstrate superior performance in certain scenarios. Furthermore, algorithm-specific nuances, such as computational efficiency and recall rates, are discussed across diverse datasets. Insights from this study provide valuable guidance for selecting suitable AKNN methods based on specific application requirements and dataset characteristics. Furthermore, potential directions for future research, including scalability improvements, algorithmic enhancements, and domain- specific applications are identified to further advance the field of AKNN algorithms.