Financial-DataMining

This repo contain codes to predict future stock price by machine learning algorithms.

Models

A series of ML models are implemented.

• Random Forest
• AdaBoost
• XGBoost
• DNNs
• CNNs
• RNNs
• Transformer

Environment

• Python 3.6
• Scikit-Learn 0.20
• XGBoost
• PyTorch1.0

Dataset

Due to credential reason, raw dataset cannot be publicised for now. Processed data may be uploaded later.

Raw data is first processed via Data_Preprocess.ipynb, which is fundamental step for whole project.

How to RUN

• For those ensemble models, simply run corresponding jupyter notebooks cell by cell

• For NN models, go to ./Regression_NN run Training.py

  Training.py has a series of parameter you can adjust. For the details you may need to read the Training.py source code.

  Here is an example for training Transformer model:

  python Training.py --batch_size 1024 --prev_step 30  --suffix pre30_class_raw --num_classes 1 --epochs 30 --feature_size 256 --normalize --layers 1  --model Transformer

Results to show (MSE metric)

Model	Random Forest	AdaBoost	XGBoost	CNN	DNN	GRU	LSTM	Transformer
Raw Data	0.555	0.601	0.546	0.551	0.554	0.548	0.549	0.545
DWT	0.512	0.587	0.481	0.495	0.488	0.489	0.485	0.482
DWT merge	0.490	0.582	0.464	0.486	0.473	0.486	0.484	0.471
SAE				0.543
Feature Synthesis	0.554	0.593	0.545

Trading Strategies:

Codes are in one of my partners' repo