/Analysis-of-Stock-High-Frequent-Data-with-LSTM

A simply framework of researching stock data through LSTM by Tensorflow

Primary LanguagePythonMIT LicenseMIT

Analysis-of-Stock-High-Frequent-Data-with-LSTM

Introduction

This project aims at predicting stock price based on high frequency stock data. There is a big difference between high frequency data and others, thus certain preprocessing methods are necessary in mining useful information. LSTM is again proved effective in this problem. As a contrast, we also tested some other classical machine learning model such as XGBoost and random forest.

Experiment

Prediction of next tick's price:

We use LSTM to predict stock price, mid-price of next tick. Random Forest and XGBoost are used to classify the following price trend.

  • label: next price delta

  • label: next mid price delta

Prediction of future mean price:

  • label: 2.5 min mean price delta

Feature importance:

The size of circle indicates its feature importance.

  • model: Random Forest, label: next price delta

  • model: XGBoost, label: next price delta