/LSTM-Stock-Predictions

Prediction of Stock price using Recurrent Neural Network (RNN) models. Contains GRU, LSTM, Bidirection LSTM & LSTM combinations with GRU units. The models were deveoped using the keras module from Tensorlfow.

Primary LanguagePython

Stock-Predictions

Prediction of Stock price using Recurrent Neural Network (RNN) models. In this project, I compare how different well-optimized RNN models perform at stocks prediction.

Models

I have used some of the most popular RNN models that are used in the today's industry:

  1. LSTM: https://arxiv.org/pdf/1909.09586.pdf
  2. GRU: https://arxiv.org/pdf/1412.3555.pdf
  3. Bidirectional LSTM: https://arxiv.org/pdf/1802.00889.pdf
  4. Proposed Method: Deep Bidirection GRU with LSTM on Output

Dataset

The source of my datasets is Yahoo's finance website: https://finance.yahoo.com/

The datasets include Google's, Tesla & Greek's Alpha-Bank stocks. Specifically, each dataset contains training data about the stocks from 01/01/2017 to 01/01/2019. Then, a small dataset from 01/01/2019 to 01/01/2020 is used to make the predictions. The dataset contains the following data for each stock:

Data | Open | High | Low | Close | Adj Close | Volume

  • Open: The inital price of the stock at the beginning of the day.
  • High: The highest price of the stock at that particular day.
  • Low: The lowest price of the stock at that particular day.
  • Close: The final price of the stock at that particular.

Libraries

The RNN were implemented using Python. The libraries that were used are the following:

  1. Numpy
  2. Pandas
  3. Matplotlib
  4. Keras
  5. Tensorflow
  6. Tensorflow Addons

Results

Google's Prediction

Tesla's Prediciton

Alpha Bank's Prediction