Application of Logistic Regression with Filter in Data Classification

Logistic regression is the main method to deal with data classification in the field of large data and machine learning. The traditional logistic regression uses gradient descent method to solve the optimal parameters of loss function with convexity to a certain extent. In the case of non-convex loss function, the filter method can replace the traditional penalty function method to ensure the global convergence of the optimization algorithm. The filter method does not need to consider the selection of parameters, thus avoiding the shortcomings of the penalty function method. The main idea of the filter method is to accept the trial point when the objective function or constraint violation degree is improved. In this paper, filter theory in the field of operations research and cybernetics is used to solve the optimal parameters of loss function, which overcomes the limitation of gradient descent method on convexity requirement of loss function. In this paper, we use logistic regression with filters to deal with the problem of actual data classification, and verify the effectiveness of the algorithm.