Elastic Network Regression Based on Differential Evolution Dragonfly Algorithm with T-Distribution Parameters

摘要

Elastic network regression in machine algorithm is a linear regression algorithm with stronger stability and wider trial range by introducing L1 and L2 regularization. The problem of collinearity cannot be dealt with well in traditional multivariate linear regression. By add in L1 and L2 regular terms, two other regressions are produced, namely Lasso regression and ridge regression. Ridge regression can avoid overfitting, but the model has poor interpretation and high complexity. Lasso regression model can reduce some regression coefficients but it will produce incomprehensible points and obvious limitations, so the emergence of elastic network regression is particularly necessary. The traditional dragonfly algorithm does not have fast convergence speed and high precision, and it is not very effective to solve the elastic network regression problem. Differential evolution dragonfly algorithm adds three core processes to the original algorithm, which increases the global optimization ability of the algorithm, increases the optimization accuracy of the algorithm, and overcomes the problems existing in the traditional algorithm. Therefore, it is advisable to use the dragonfly algorithm of differential evolution to solve the elastic network regression.