Applying machine learning to stock market trading

Abstract

With the development of the artificial intelligence, the application of machine learning in finance field has attracted extensive attention from investors and researchers. This article combines stock selection, stock data pre-processing, neural network LSTM improved model to predict stocks. We first used three companies' financial indicators (Cash flow per share/earnings per share; Return on net assets; Sales gross margins) to analyze the 3,000 stocks in Shanghai. Results showed that with the minimum clustering principle, choose a stock is more valuable to the investment. Next, we used the quantitative trading platform to collect six indicators of stock data: the opening price, the closing price, the highest price, the lowest price, volume and turnover. Then six feature data were preprocessed using wavelet denoising. We further used the modified three-layer LSTM model to predict the closing price of the transaction with denoising data. This article analyzes the influence of the scale of the training data on the prediction results and finds that the de-noise processing has a better prediction effect in the small-scale data model. Meanwhile, the prediction result is better when the learning rate is 0.004. Overall, the LSTM neural network has a better prediction result with denoise shortterm data than the original long-term data.