Framework Design for Demand Forecasting Using Time Series Decomposition

This technical report presents a framework for demand forecasting that utilizes a time series decomposition-based approach to enhance both accuracy and interpretability. The study emphasizes the importance of not only achieving high accuracy in demand forecasts but also providing a clear basis for decision-making. It introduces the Seasonal-trend decomposition using locally estimated scatterplot smoothing (STL) method, which decomposes time series data into trend, seasonality, and residual components. The ARIMA model is employed to forecast trends and residuals, resulting in a model that maintains high accuracy while improving interpretability compared to traditional methods like SARIMA. The report outlines the methodology for implementing the STL-ARIMA framework, detailing the steps involved in time series decomposition and the subsequent forecasting process. It highlights the significance of interpretability in demand forecasting, particularly in business contexts where decision-makers require a clear understanding of forecast results to make informed choices regarding inventory management and supply chain operations.