bodhi12496/stock_prediction

PREDICTING STOCK PRICE USING LSTM

This report presents our project on forecasting stock prices using a feature fusion LSTM-CNN model based on the research paper "Forecasting Stock Prices with a Feature Fusion LSTM-CNN Model Using Different Representations of the Same Data." We aim to replicate the paper's model and explore the impact of different data representations on stock price prediction.

Tech Stack

Python , LSTM , Sc-CNN model , Tensorflow , Keras

Results obtained on implementing the models:

LSTM Model:

• Loss: 1.2461
• MAPE: 5.43%
• RMSE: 1.1163

SC-CNN Model:

• Loss: 0.0416
• MAPE: 5.44%
• RMSE: 0.2040

Fusion Model:

• Loss: 39900.3945

Recommendations to improve the model performance:

• Investigate and address issues causing high loss in the Fusion model.
• Explore alternative fusion techniques and architectures.
• Consider additional feature engineering and preprocessing steps for improved model performance.

Few modifications performed on the fusion model:

• Additional Dense Layers: The code adds more dense layers to the Fusion model, allowing it to capture more complex relationships.
• Learning Rate Adjustment: Lowering the learning rate can sometimes help the model converge more effectively and avoid overshooting the minimum.
• Different Activation Functions: Adjusting activation functions in the additional layers might improve the model's ability to capture non-linear patterns.
• Training with Adjusted Parameters: The model is then trained with the adjusted architecture and hyperparameters

Results after fine tuning fusion model:

The following are the values of the fusion loss obtained on the different representations with and without performing modifications Without Refinement:

• Fusion Model on Stock and Linebar: 12075.2832
• Fusion Model on Stock and F-linebar: 12075.2832
• Fusion Model on Candlebar: 12075.2832
• Overall Fusion Model: 39900.3945 With Refinement :
• Fusion Model on Stock and Linebar: 9658.22
• Fusion Model on Stock and F-linebar: 9658.22
• Fusion Model on Candlebar: 9658.22
• Overall Fusion Model: 31920.3156