README.md

metadata

title: DataSynthis ML JobTask
emoji: 📈
colorFrom: blue
colorTo: green
sdk: streamlit
sdk_version: 1.28.0
app_file: app.py
pinned: false
license: mit

DataSynthis ML JobTask: Stock Price Forecasting

This application demonstrates comprehensive stock price forecasting using both traditional statistical methods (ARIMA) and modern deep learning approaches (LSTM).

🎯 Features

• Dual Model Approach: Compare ARIMA and LSTM forecasting models
• Interactive Interface: Upload your own stock data and get instant predictions
• Comprehensive Analysis: Rolling window evaluation and performance metrics
• Visual Comparisons: Professional charts and forecasting visualizations
• Export Capabilities: Download forecast results as CSV

📊 Models Included

1. ARIMA (AutoRegressive Integrated Moving Average)

• Traditional statistical time series model
• Excellent for capturing linear trends and seasonality
• Fast training and interpretable results
• Optimal for shorter-term predictions

2. LSTM (Long Short-Term Memory)

• Deep learning neural network architecture
• Capable of learning complex non-linear patterns
• Handles long-term dependencies in time series
• Powerful for capturing market behavior nuances

🚀 Usage

1. Upload Data: Upload your stock price CSV file with required columns
2. Select Model: Choose between ARIMA, LSTM, or both models
3. Configure: Set forecast horizon (1-60 days)
4. Generate: Click to train models and generate predictions
5. Analyze: View comprehensive results and download forecasts

📁 Expected Data Format

Your CSV file should contain the following columns:

• date: Date in YYYY-MM-DD format
• open: Opening price
• high: Highest price of the day
• low: Lowest price of the day
• close: Closing price
• volume: Trading volume
• Name: Stock symbol/company name

🧪 Model Performance

Based on comprehensive testing with Apple (AAPL) stock data:

Model	RMSE	MAE	MAPE (%)	R-squared
ARIMA	$4.99	$3.65	3.13%	0.960
LSTM	$6.41	$5.00	4.23%	0.934

ARIMA showed statistically significant better performance (p < 0.05) in rolling window evaluation.

🔬 Technical Implementation

• Rolling Window Evaluation: Robust backtesting methodology
• Statistical Significance Testing: T-tests for model comparison
• Comprehensive Metrics: RMSE, MAE, MAPE, Direction Accuracy, R-squared
• Professional Visualizations: Multiple chart types for thorough analysis

🛠️ Built With

• Streamlit: Interactive web application framework
• Statsmodels: ARIMA implementation and statistical analysis
• TensorFlow/Keras: LSTM neural network implementation
• Pandas & NumPy: Data manipulation and numerical computing
• Matplotlib & Seaborn: Data visualization
• Scikit-learn: Machine learning utilities and metrics

📈 Use Cases

• Financial Analysis: Stock price prediction and trend analysis
• Investment Strategy: Support decision-making with quantitative forecasts
• Risk Management: Understand model uncertainties and confidence intervals
• Educational: Learn about different forecasting approaches
• Research: Compare traditional vs. modern ML approaches

🎓 Key Learnings

This project demonstrates that:

• Complexity ≠ Performance: Simpler models (ARIMA) can outperform complex ones (LSTM) for certain problems
• Domain Knowledge Matters: Understanding time series characteristics is crucial
• Proper Evaluation: Rolling window validation provides realistic performance estimates
• Model Selection: Choose approaches that match your data's inherent complexity

📜 License

MIT License - feel free to use this code for your own projects!

🤝 Contributing

Contributions, issues, and feature requests are welcome! This project was created as a demonstration of comprehensive ML model comparison and deployment.

---

Built by DataSynthis Team - Showcasing the power of combining traditional statistics with modern machine learning for financial forecasting.