Managing missing data and predictions in short time series

António, Francisco de Araújo; Cavique, Luís

http://hdl.handle.net/10400.2/22008

Utilize este identificador para referenciar este registo.

Nome:	Descrição:	Tamanho:	Formato:
1 paper EPIA missing data.pdf		1.25 MB	Adobe PDF	Ver/Abrir

Contacte-nos

Autores

António, Francisco de Araújo

Cavique, Luís

Resumo(s)

Sales forecasting in the presence of Missing Data poses significant challenges, particularly for short time series where limited observations amplify the impact of incomplete records. This study analyzes a real-world transactional dataset (2021–2024) to predict quantities and prices for 2025. We classify miss-ingness patterns and mechanisms (MCAR, MAR, MNAR) to inform the selection of imputation strategies. We evaluate techniques including MICE, Mean, KNN, and Linear Regression under simulated missingness rates, with KNN emerging as the most robust for the MAR mechanism. Regarding very short-term series pre-dictions, the naive forecast Max2 (maximum of the last two observed values) out-performed moving averages. The results highlight the importance of mechanism-aware imputation and domain-tailored forecasting in sparse datasets. This work presents a practical framework for businesses to effectively utilize incomplete sales data.