Research on the Learning Method Based on PCA-ELM

Abstract

The Single-hidden Layer Feed-forward Neural Network has been widely applied in the fields such as pattern recognition, automatic control and data mining. However, the speed of the traditional learning method, since it is far from enough to satisfy the actual demand has become the main bottleneck, which restricts its development. As one of the new learning methods, the extreme learning machine (ELM) has its own remarkable characteristics, but the fact that ELM is based on the Empirical Risk Minimization may lead to over fitting. In addition, ELM does not consider the weight of error, so its performance will be severely affected when there are outliers in data integration. To solve the above problems, this paper referred to the two algorithms including PCA (Principal Component Analysis) and ELM, and put forward a learning method and prediction model, which combined PCA and ELM. From the results of simulation analysis, as combining advantages of PCA and ELM algorithms, the network structure can be simplified to improve the learning ability and its prediction precision.

KeyWords:

• Extreme learning machine (ELM)
• Principal component analysis (PCA)
• Feed-forward neural network