Impact of Covid-19 on Demand Planning: Building Resilient Forecasting Models

Ashok, S.; Aravind, K.

Early predictions are crucial in helping organizations to stay stable and to be well positioned. Forecasting models have a fundamental presumption that the historical data offers all dimensions needed to predict the seasonality, cyclic patterns and trends with a reasonable degree of uncertainty. Nevertheless, with regards to demand, such hypothesis is being challenged by the huge impact of Covid-19 on business. The pandemic has significantly altered the spending patterns of the customer and immensely influenced business in major industries like travel and hospitality. Many organizations are not able to rely on their forecasting solutions for revenue estimates that are critical to cost projections, inventory and planning due to the economic uncertainties, evolving conditions and irregular data spikes. In this research, we focus on various techniques to handle the data abnormalities and recommends a decomposition-based data curation process for reliable forecasts using change point detection and also utilizing the characteristics of both pandemic and recovery periods. In order to gauge the predictive performance of the proposed approach, the results are compared with a pool of imputation techniques and forecasts derived from autoregressive, machine learning and deep learning models. The proposed approach generates accurate imputations on real world datasets, especially those having highly seasonal patterns. © 2021 ACM.