A Dimension Reduction Method Based on Adaptive RBF Neural Network in Supervised Machine Learning

Abstract

For the purpose of high dimensional data reduction in supervised machine learning, an adaptive RBF (Radial Basis Function) neural network algorithm was proposed. According to the relationship between the total error of this training and last training, the learning rate and momentum factor in the algorithm were adjusted dynamically and adaptively. So, the learning direction of RBF neural network was enhanced and the learning speed was improved. This algorithm was used to reduce the dimension of high dimensional data. The data set auto-mpg in UCI machine learning database was taken as an example to carry out dimension reduction experiments. The experimental results show that the learning performance of RBF neural network was improved and its general error was reduced after reducing the dimensionality of high dimensional data set by this algorithm. This reduction method is of great significance to solve the “dimensionality disaster” and “cost disaster” in machine learning.