Multi-Class Logistic Regression by Randomization and Nonlinearization

Key Points

• Improved accuracy in multi-class logistic regression results from randomization and nonlinearization techniques, enhancing data interpretability.
• In specific numeric examples, the joint effects of high-dimensional randomization and nonlinearization alter the feature space characteristics significantly.
• Verification was conducted using metrics like matrix rank and eigenvalue distribution, indicating desirable properties in the feature space were achieved.
• Indicating a potential shift in how multi-class models can be effectively structured in complex data environments.

Abstract