Research on Dimension Reduction Algorithm of Prediction Model Input Variables Based on Principal Component Analysis

Abstract

The selection of input variables of neural network prediction model is very important. If there are too many input variables, the input layer nodes of the network will increase and we need an exponential number of training samples which can easily lead to “disaster of dimension”. According to the characteristics of so many influencing factors of cervical spinal cord injury and intertrochanteric fracture, principal component analysis is used to screen the input variables of neural network prediction model in this paper. And in this method, the principal component factor with the greatest contribution is selected as the input node of the neural network. Then it not only comprehensively considers the various information contained in the input variables, but also compresses the number of input variables to the greatest extent, so as to simplify the topology of the neural network. At the same time, the training time is greatly shortened. Simulation results show that the method is effective and feasible.