Project 6: Time Series Forecasting

Forecast future values in a time series (sales, stock prices, etc.).

Pick one path: This project overlaps with Module 15 and advanced project 3. See TIME_SERIES_LEARNING_PATH.md.

Starter code: Run starter.py after placing data in data/.

Difficulty

Intermediate

Time Estimate

4-5 days

Skills You'll Practice

• Time Series Analysis
• ARIMA Models
• LSTM Networks
• Feature Engineering for Time Series

Learning Objectives

By completing this project, you will learn to:

• Preprocess time series data
• Identify trends and seasonality
• Apply ARIMA models
• Use LSTM for sequences
• Engineer time-based features
• Evaluate time series models

Dataset

Airline Passengers Dataset

• Kaggle Airline Passengers
• Monthly passenger numbers
• Clear trend and seasonality
• Good for learning

Project Steps

Step 1: Load and Explore Data

• Load time series data
• Visualize time series
• Check for missing values
• Identify trends and seasonality

Step 2: Time Series Decomposition

• Decompose into trend, seasonality, residual
• Visualize components
• Understand patterns

Step 3: Stationarity

• Check for stationarity (ADF test)
• Apply differencing if needed
• Make series stationary

Step 4: ARIMA Model

• Identify ARIMA parameters (p, d, q)
• Use auto_arima or manual selection
• Train ARIMA model
• Forecast future values

Step 5: LSTM Model

• Prepare data for LSTM
• Create sequences
• Build LSTM network
• Train and forecast

Step 6: Feature Engineering

• Create lag features
• Add time-based features (month, day, etc.)
• Create rolling statistics
• Use external features if available

Step 7: Model Evaluation

• Split into train/test (time-based!)
• Calculate RMSE, MAE
• Visualize predictions
• Compare ARIMA vs LSTM

Expected Deliverables

1. Jupyter Notebook with complete analysis
2. Forecasted Values for future periods
3. Visualizations of predictions vs actual
4. Comparison Report of different methods

Evaluation Metrics

• RMSE: Root Mean Squared Error
• MAE: Mean Absolute Error
• MAPE: Mean Absolute Percentage Error
• Visual Inspection: Plot predictions

Models to Implement

1. ARIMA: Classical time series
2. LSTM: Deep learning approach
3. Prophet: Facebook's tool (optional)
4. Simple Methods: Moving average, exponential smoothing

Tips

• Use time-based train/test split (not random!)
• Check for stationarity
• Handle seasonality properly
• Create lag features
• Visualize everything
• Compare multiple approaches

Resources

• Kaggle Time Series
• ARIMA Guide
• LSTM for Time Series

Extensions

• Multiple time series
• External features
• Real-time forecasting
• Confidence intervals
• Anomaly detection in time series

Next Steps

After completing this project:

• Try more complex time series
• Experiment with advanced techniques
• Move to Advanced Projects