Apple Stock ML Project

A machine learning project for predicting extreme events in Apple stock prices using Random Forest and Temporal CNN models.

Installation

Prerequisites

• Python 3.10 or higher
• Poetry (Python package manager)

Setup

1. Install Poetry (if not already installed)

2. Clone the repository

git clone https://github.com/mchaniotakis/apple_stock_ml

cd apple_stock_ml

3. Install dependencies using Poetry

poetry lock
poetry install

Project Structure

src/
├── data_processing.py # Data loading and preprocessing functions
├── random_forest.py # Random Forest model implementation
├── temporal_cnn.py # Temporal CNN model implementation
├── model_evaluation.py # Model evaluation and metrics calculation
└── improvement.py # Model improvements and enhancements

Data

• You dont have to specifically download any data, all models use data_processing to import the data requested.
• For every model you can specify the --ticker stock_name option to use that stock for training. Every stock is cashed so that you dont have to download it twice.
• You can download multiple stocks using SP500 for the S&P500 stocks, or if you would like the top X stocks, you can use this ticker : SP500TOP10 which will download the top 10 stocks and use them for training.
• Make sure to change the --start-date --end-date if you need to use different dates.
• Use --threshold to specify the threshold for the extreme event point, currently at 2.0
• --model-output specifies the output location of the model saved after its trained, currently at models/

Evaluation

• Once the models are trained, they would all be located at models/.. at that point feel free to run the evaluation script to compare their performance. Please take a look at evaluation_results
• Run with python apple_stock_ml/src/model_validation.py --ticker AAPL --rf_idx_to_keep [-1] which makes sure that the RF model is only using the daily return feature.
• The performance metrics for both models are stored in the evaluation_results directory, including: - Confusion Matrix - Accuracy - Precision, Recall, F1-Score - ROC curves

Train models

Random Forest Model

Run the Random Forest model with default settings:

python apple_stock_ml/src/random_forest.py --ticker AAPL

Optional flags:

--use-smote: Enable SMOTE for handling class imbalance

-o: Enable optimization mode for hyperparameter tuning

Temporal CNN model

1. run the Temporal CNN model with the default settings:

python apple_stock_ml/src/temporal_cnn.py --ticker AAPL --learning-rate 0.1 --epochs 300 --patience 15 --batch-size 256

2. Optional flags:

--pca: Enable PCA dimensionality reduction

--sequence-length: Set custom sequence length (default: 10)

--augs: Enable data augmentation

--use-smote: Enable SMOTE for handling class imbalance

Improved Model

1. Run the improved model with the default settings:

python apple_stock_ml/src/improvement.py --ticker AAPL --sequence-length 10 --learning-rate 0.01 --batch-size 256 --patience 15

2. Optional flags:

--pca: Enable PCA dimensionality reduction

--sequence-length: Set custom sequence length (default: 10)

--augs: Enable data augmentation

--use-smote: Enable SMOTE for handling class imbalance

Model Validation

1.Run the model validation (we are using only the dialy return for RF):

python apple_stock_ml/src/model_validation.py --ticker AAPL --rf_idx_to_keep [-1]

The validation script will:

• Load the trained models (from models/...pt)
• Generate performance metrics
• Create confusion matrices
• Compare model performances

Reports and Documentation

• Detailed analysis and results can be found in report.pdf
• Model architecture details and hyperparameters are documented in the respective source files
• Performance comparisons and improvement strategies are detailed in the report

Notes

• All models will save their outputs in the appropriate directories under the project folder
• Results and metrics will be stored in the evaluation_results directory
• Models can be run with different stock tickers by changing the --ticker parameter
• Use the --help flag with any script to see all available options
• For more details about the implementation and metrics, refer to the project documentation and the original assignment requirements.