Time Series Forecasting with Holt-Winters (Python)

This project demonstrates how to build a time series forecasting model using the Holt-Winters Exponential Smoothing technique on a synthetic monthly sales dataset. The goal is to model and predict future values using trend and seasonality.

Overview

• Generates 5 years (60 months) of synthetic monthly sales data with trend, seasonality, and noise
• Trains a Holt-Winters (Triple Exponential Smoothing) model using statsmodels
• Forecasts the next 12 months of sales
• Evaluates forecast accuracy using Mean Absolute Error (MAE)
• Visualizes training data, actual values, and forecasted values on one plot

Results

• Mean Absolute Error (MAE): 8.13

Key Insight

The Holt-Winters model successfully captured both the upward trend and the seasonal monthly pattern in the data, producing a forecast that closely tracks actual sales — visible in the plot where the forecasted line (red) stays close to the actual test values (green).

Tech Stack

Python, Pandas, NumPy, Matplotlib, Statsmodels, Scikit-learn

How to Run

pip install -r requirements.txt
python Time_Series_forecasting.py

Project Structure

Time-Series-Forecasting/ │ ├── Time_Series_forecasting.py # Generates data, trains model, forecasts, plots results ├── requirements.txt └── README.md

Future Improvements

• Test on a real-world sales dataset instead of synthetic data
• Compare against other forecasting methods (ARIMA, SARIMA, Prophet)
• Add cross-validation across multiple forecast windows
• Deploy as an interactive Streamlit demo