DIMENSIONALITY REDUCTION AND FITTING METHOD FOR HIGH-DIMENSIONAL DATA BASED ON SVD AND LEAST SQUARES—A CASE STUDY OF MINE DATA PROCESSING

In the digital era, the explosive growth of high-dimensional data poses significant challenges to storage, transmission, and computational efficiency. Mine data, characterized by its multi-source heterogeneity, high dynamism, and high dimensionality, presents particularly acute challenges. This paper proposes a novel method for dimensionality reduction and fitting of high-dimensional data by combining Singular Value Decomposition (SVD) with least squares, and demonstrates its first application in mine data processing. The method achieves efficient data compression and precise fitting by extracting principal singular values and vectors via SVD, projecting high-dimensional data into a low-dimensional space, and solving for optimal weight vectors using least squares. A pseudo-inverse is constructed to avoid numerical instability, ultimately completing the fitting of the target dataset. Experimental results show that the method performs exceptionally well in terms of residual distribution, model bias, data noise, and fitting adequacy: residuals approximate a normal distribution, confirming that errors primarily stem from data noise. This study provides a reliable technical pathway for processing high-dimensional mine data, with future optimizations possible through the introduction of noise reduction modules.