Algorithms for stock's next n days up/down movement classification.
- Python3
- Dependecies are in requirements.yml
You may train using one of the available algorithms (note that hyperparameters have not been optimized and must be adjusted by yourself):
150 input features (open, high, low, close, volume for the last 30 days)
Neural Network: CNN with LeakyReLU in hidden layers and softmax in output layers, regularized by dropout and batch normalization, Adam as an optimizer.
SVM: C = 100, the rest are default from sklearn.
Random Forest: number of trees = 25, the rest are default from sklearn.