Interpretability Analysis Of Stock Price Forecasting Based On Machine Learning Models

The stock price forecasting model is used to predict the future price trend of stocks by current information,with which investors can make good investment decisions.As the stock price is affected by many factors,the model with simple structure can hardly complete the task of stock price forecasting.Even though the model with high complexity can achieve good prediction performance,its poor interpretability compresses its upper limit and limits its further application and development.To solve this problem,we select a variety of machine learning models as alternative models,and use related data to train stock price forecasting models respectively.We improve the existing machine learning interpretation(LIME algorithm)for stock data,and use the improved interpretation method to partially explain the stock price forecasting model with the best prediction performance of alternative models.The result shows that Light GBM achieves the best performance in the stock price forecasting problem set in this paper.For this model,the improved interpretation method in this paper has achieved great in terms of fidelity,consistency and comprehensibility,and is significantly better than the unmodified LIME model.The improved interpretation method in this paper is universal in the problems of stock related models.We hope that the interpretation method can not only act on the problems set in this paper,but also on other stock-related problems,and promote the development and application of stock-related models from the perspective of model interpretation.