Research on Stock Index Prediction Based on Stock Correlation Network and Deep Learning

Aiming at the problem of stock index prediction, constructing a time series correlation network based on fundamentals and technology components, then using depth map neural to learn hierarchical representation network, obtaining candidate prediction signal in an end-to-end way. The architecture composed method strategy is called DIFFPOOL architecture. Taking CSI 300 as research object, combining with softmax classifier, long-term short-term memory (LSTM), linear regression, logical respectively, uses sliding window obtain corresponding accuracy index. combined model under optimal parameters fluctuates interval [0.56, 0.62]. Ultimately, first mock exam mean absolute error (MAE) root square (RMSE). regression models are compared LSTM, recurrent (RNN), back propagation (BP). Compared single model, MAE RMSE smaller, 0.0061 0.0081, respectively. Experiments show that by aggregating node attribute information association hierarchically, we can dynamically capture impact different industry sectors price fluctuations further improve accuracy.