Improvement of Multi-Class Logistic Regression by Dimensional Change

In this paper, a method to improve the accuracy of multi-class logistic regression analysis is proposed. Characteristics of the data matrix from the perspective of numerical analysis is examined. In addition, eigenvalue distribution of the data matrix is considered, demonstrating that the number of large eigenvalues, serving as important signals, varies depending on the dimension of randomization.