Time Series Forecasting

I developed a comprehensive Time Series Forecasting system using the Darts framework in Python, designed to model and predict sales trends for a retail store. The project focused on experimenting with a wide range of forecasting techniques to identify the most accurate and efficient approaches for different data scenarios.

Statistical / Classic Models • ARIMA • ⁠Exponential Smoothing • ⁠Prophet

Regression Models • Linear Regression • ⁠Random Forest • ⁠LightGBM • ⁠XGBoost • ⁠CatBoost

PyTorch (Lightning) Based Models • N-HiTS • ⁠TCN • ⁠Transformer • ⁠D-Linear • ⁠N-Linear • ⁠TiDE • ⁠TSMixer

By leveraging traditional statistical methods, machine learning regressors, and state-of-the-art deep learning architectures, this project highlights the scalability of Darts for handling complex forecasting tasks. The outcome was a flexible forecasting pipeline capable of adapting to varying demand patterns, ensuring more accurate, data-driven decision-making for inventory and business planning.