Study on BP Neural Network Fitting for High-Dimensional Small-Sample Data Based on PCA Dimensionality Reduction

This paper studies the preprocessing method for high-dimensional small-sample data. High-dimensional small-sample data usually has feature redundancy and multicollinearity, which makes the generalization ability of prediction models worse. This paper uses principal component analysis (PCA) to reduce data dimension. It selects useful principal components by variance contribution rate and uses data reconstruction to check whether key information is kept. A BP neural network with fixed parameters is used for comparison tests. The results show that the test-set coefficient of determination rises from 0.94671 to 0.9715 after PCA processing. It proves that PCA can remove redundant information and improve input data quality for high-dimensional small-sample fitting.