Wavelet Neural Network Model with Time-Frequency Analysis for Accurate Share Prices Prediction

Due to the large amounts of risks and potential financial benefits involved, ability achieve accurate prediction on stock market prices is great interest investors. However, non-stationarity, high level volatility, frequent fluctuations stochastic properties that data possesses, have made it difficult accurately predict share prices, even by recently developed deep learning methods. This can be attributed outputs trained are not responsive enough capture rapid adjustments in real data, hence affecting accuracy. To solve these difficulties, this paper proposes a wavelet neural network model using Gaussian as activation function decomposing into finer precision with account for sensitivity, further optimising mapping process detailed time-frequency analysis outputs, leading higher accuracy faster speed. The proposed two training processes has been validated dataset from London market, results demonstrated model-based predictions distinctly superior current methods, which corresponds significant reduction mean squared error.