Construction Of Stock Evaluation Index System And Prediction Method Based On Artificial Intelligenc

As a core component of the capital market,the price trend and future trend prediction of the stock market is one of the most important concerns for investors.Accurately predicting stock prices can provide investors with strong technical support.In recent years,deep learning has been widely used in stock prediction to explore more effective information.However,the nonlinear,multiscale,and complexity characteristics of stock data make it relatively difficult to extract hidden information.Deep learning prediction models also face problems such as gradient disappearance and time lag,making it difficult to fit when there are large sequence fluctuations,and the prediction accuracy needs to be improved.To address the aforementioned issues,this paper attempts to establish and select a scientific method for stock evaluation technical indicators,and improve stock prediction accuracy by combining the Broad Learning System(BLS)with the Long-Short Term Memory(LSTM)network in deep learning.The main work and innovative points are as follows.(1)For the single-input stock prediction problem considering only the closing price,the CEEMD-LSTM-BLS prediction model is proposed: The model uses Complementary Ensemble Empirical Mode Decomposition(CEEMD)to perform non-stationary preprocessing on the closing price.To improve the time lag phenomenon commonly found in deep learning in stock prediction,we attempt to combine broad learning with deep learning and propose the LSTMBLS hybrid prediction module.(2)Considering the problem of insufficient information in the closing price,the introduction of technical indicators is necessary.However,the existing number of stock technical indicators is numerous and lacks standardized selection criteria.This paper attempts to scientifically and reasonably construct a technical indicator system for stock prediction and proposes an intelligent selection method for stock evaluation technical indicators based on rough set theory(RSS).(3)The Fuzzy BLS(FBLS)is used to improve the generalization ability and robustness of BLS.The technical indicators selected in(2)are combined to propose a multi-factor stock price prediction model based on RSS and CEEMD-LSTM-FBLS.Empirical research is conducted on four representative stock index data sets: Shanghai Composite Index,Shenzhen Component Index,Dow Jones Industrial Average,and S&P 500.The research results show that:(1)Compared with existing single-input prediction models,the CEEMD-LSTM-BLS model proposed in this paper has the best prediction performance on evaluation indicators,but there is still room for improvement in the prediction accuracy for two volatile foreign data sets;(2)Based on RSS,a technical indicator system for stock prediction can be constructed reasonably and effectively.Compared with the input of only basic and technical indicators,the prediction under this system takes into account both the control of training time and the improvement of prediction accuracy;(3)Reasonable construction of technical indicators and the use of improved FBLS further improve the model’s prediction accuracy.In particular,the determination coefficients for foreign data prediction can also reach0.9805 and 0.986.The proposed prediction model has certain generalization ability and accuracy and provides valuable reference for stock price prediction.