Multicollinearity Diagnostics on Cyclostationary Features
Abstract
Cyclostationary features such as high-order moments and cumulants are often applied in detecting and classifying digitally modulated signals. However, these predictor variables may not be mutually uncorrelated, raising the concern that potential multicollinearity may lead to redundant information and significantly affect quality prediction. This report explores machine learning and statistical methods for detecting and mitigating multicollinearity in cyclostationary features. Our study examines features corresponding to synthetic electromagnetic waveforms of nine modulation types over nine levels of signal-to-noise. The empirical results indicate that the Variance Inflation Factor (VIF) method and the Eigendecomposition of the predictor correlation matrix are effective in detecting multicollinearity. Furthermore, employing feature transformation using the Principal Component Analysis (PCA) technique allows us to identify principal components that are less influential in the performance of prediction models.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 07, 2022
- Accession Number
- AD1184366
Entities
People
- Anthony S. Tai
Organizations
- Naval Surface Warfare Center Crane Division