Comparative Analysis of Feature Extraction of High Dimensional Data Reduction Using Machine Learning Techniques
Seth Gyamerah, Godfred Tour Soori, Dennis Redeemer Korda, John Kwame Tawiah, Eric Ayintareba Akolgo, Emmanuel Oteng Dapaah · American Journal of Electrical and Computer Engineering · 2023
Dimensionality reduction is critical for analyzing and interpreting high-dimensional data across domains like genomics, imaging, and finance. This paper presents a comparative analysis of dimensionality reduction techniques, including Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), Recursive Feature Elimination (RFE), and Lasso regression. These methods are applied to datasets from genomics, medical imaging, and finance to evaluate their ability to reduce dimensions while preserving relevant information. The results demonstrate that PCA and LDA are highly effective for genomics data, reducing gene expression profiles from over 60,000 dimensions to 10-50 components while maintaining precision of over 80%. For medical images, PCA and LDA reduce pixel dimensions by over 90% without compromising precision. However, no single technique optimizes dimensionality reduction and precision for complex finance data. Overall, the analysis provides domain-specific insights, highlighting PCA and LDA as leading techniques for genomics and imaging. The choice of method should be guided by data characteristics. Testing on more diverse, real-world datasets is needed to establish validity further. This research aims to inform the selection of appropriate data reduction techniques across critical applications involving high-dimensional data.