Back to all cases
CASE STUDY / AI & Data Science
Disrupted Time-Series Forecasting under COVID-19
MarTech, Retail & Demand ForecastingCompared SARIMA, Holt-Winters, and Prophet under COVID-19 disruptions. Designed three experimental groups (excluding / including / post-COVID data) to quantify the impact of external shocks on time-series forecasting.
Delivery Details
This project investigates how COVID-19 disruptions affect time-series forecasting accuracy. Through rigorous experimental design, data was split into three groups: excluding COVID, including COVID, and post-COVID. SARIMA (with auto-parameter optimization), Holt-Winters (triple exponential smoothing), and Prophet (automatic holiday and outlier handling) were applied to each group, comparing MAE, RMSE, and MAPE across scenarios and proposing optimal forecasting strategies for external disruptions.
Industry Context
Marketing and retail forecasts depend on historical seasonality, campaigns, and market trends, but structural shocks such as a pandemic break established patterns. The industry needs not only the lowest-error model but an interpretable strategy for choosing methods across data windows and disruption regimes.
Business Problem
External shocks distorted historical demand signals, making existing forecasting assumptions unreliable.
What I Built
Compared SARIMA, Holt-Winters, and Prophet across excluding, including, and post-shock experiments.
Business Impact
Produced interpretable model-selection guidance for retail, supply chain, and market planning.
For career conversations, advisory work, or a concrete Data & AI project, feel free to get in touch.