The ridge method in a radial basis function neural network

In this paper, the Ridge method is applied to improve radial basis function neural network. The resulting redesigned radial basis function is built to test the statistical significance in an array of independent variables considering the existence of a collinearity problem, as well as obtaining appropriate assumptions for concluding those significances. The radial basis function allows the determination of a relationship between a response, and one or more independent variables, determining the importance of each factor for the model. However, this testing may obtain negative results, when one or more columns of the design matrix are linearly dependent; for this reason, we have adapted the Ridge method for the radial basis function. The results show that the variance inflation factor is a good metric alternative for validating the effectiveness of neural network inference. Our primary conclusion is that the redesigned radial basis function results in improved model accuracy when combined with the Ridge method. Additionally, this model can also be used to validate the statistical assumptions required to find the sources of the multicollinearity in an analysis, discovering the corrections and interpreting the model.