A Novel Prediction Method based on Grey-LVQ Neural Network

This paper proposes a prediction method that combines a grey system, Fourier series, and neural network in order to accurately predict a time series. The proposed method is referred to as GFLV QGRG (Grey Fourier LVQ based on GRG). The grey model is first used in combination with a Fourier residual correction to forecast future stock indices. A learning vector quantization (LVQ) neural network and Self-Organizing Map (SOM) neural network are utilized to conduct clustering analysis on the magnitude of fluctuation in the stock indices, resulting in clusters with different volatilities. Residual correction based on the Fourier sine or cosine series is applied to the clusters to obtain the most optimal prediction. Our experimental results show that compared with traditional time series prediction models GFLV QGRG has superior accuracy.