NON-LINEAR HYBRID RANDOM PROJECTION TECHNIQUES FOR HIGH-DIMENSIONAL DATA

Abstract

Dimensionality reduction of large datasets is essential in fields such as cloud computing, genetic distribution, image processing, and smart grids, which require high-dimensional data analysis. Despite their efficiency, Linear Hybrid Random Projection methods are limited in capturing the inherent structure of complex nonlinear data. This limitation necessitates the research to introduce a novel Nonlinear Hybrid Random Projection (NHRP) method, combining normal random projection (NRP) and plus-minus random projection (PMRP) with nonlinear transformations. This approach aims to enhance the representation of nonlinear relationships, thereby improving the accuracy and effectiveness of dimensionality reduction in various applications such as cloud computing, genetic distribution, image processing, and smart grids. NHRP dynamically balances NRP and PMRP contributions using a blending parameter, effectively reducing dimensions while preserving complex relationships. The effectiveness of NHRP is evaluated through simulations and real-world data analyses, including datasets from human activity recognition<del cite="mailto:Ridwan%20Sanusi" datetime="2024-10-09T10:45"> (HAR)</del><del cite="mailto:Ridwan%20Sanusi" datetime="2024-10-09T10:22">,</del> and hyperspectral images. Comparative studies with NRP and PMRP show that NHRP consistently outperforms in terms of distance distortion, preserving data structure across various datasets and settings. NHRP represents a significant advancement in dimensionality reduction, minimizing information loss and enhancing efficiency and precision in machine learning, intelligent systems, bioinformatics, remote sensing, and artificial intelligence.

DOI Semantic Scholar

View source