Tesla Stock Prediction Using Recurrent Nerual Network and Advanced Recurrent Neural Network Architectures

Workflow

• Data Insetion, Preparation and Visualization
  • Basic data quality check
  • Feature Engineering
  • EDA

• Model Training

  • Data Preprocessing
  
  • Created 3 different window size to train the model.
  • Trained Models using Random Search for 3 different horizons 1,5 and 10 day predictions
    • 30 Day Sequence
    • 60 Day Sequence
    • 90 Day Sequence
  • Baseline Models Stuggled in certain cases, so created an hybrid model with grid search on 30 and 60 day Sequence.
    • 30 Day Sequence
    • 60 Day Sequence

• Model Evalutaion

  • Compared all models with train,validation and test data using r2 score,RMSE,MAE and MAPE.
  • Decided which is best model for each 1 Day, 5 Day and 10 Day Horizon.

    • 1 Day Horizon – GRU (30 days Sequence)

      • Best Hyperparameters: {'gru_units': 128, 'batchnorm_1': False, 'dropout_1': 0.2, 'batchnorm_2': False, 'dropout_2': 0.5, 'learning_rate': 0.005}

        

    • 5 Day Horizon - GRU (60days Sequence)

      • Best Hyperparameters: {'gru5_units': 32, 'batchnorm_1': False, 'dropout_1': 0.5, 'batchnorm_2': False, 'dropout_2': 0.5, 'learning_rate': 0.01}

        

    • 10-day Horizon – GRU (60 days Sequence)

      • Best Hyperparameters: {'gru10_units': 64, 'batchnorm_1': False, 'dropout_1': 0.3, 'batchnorm_2': False, 'dropout_2': 0.2, 'learning_rate': 0.01}