Long-Range Forecasting of U.S. FMCPG Category Market Share Using Traditional Statistical and Machine Learning Approaches

0

Summary:
Following the end of COVID-related stimulus in March 2021, U.S. snacking growth stagnated, prompting our sponsor company to focus on capturing market share through an unbiased forecasting model that incorporates endogenous and exogenous factors to predict monthly snacking market share. The research employs a comprehensive approach, encompassing data collection from various sources, rigorous data cleaning, development and evaluation of multiple forecasting models (including traditional statistical and machine learning methods), feature importance analysis using Shapley values, and model selection based on quantitative and qualitative criteria, culminating in the generation of forecasts and their visualization through a dashboard tool. Initial analysis reveals that the SARIMAX model, which incorporates endogenous promotional factors, achieves the highest forecasting accuracy with a MAPE of 0.68%.