Original scientific paper
https://doi.org/10.17535/crorr.2026.0021
Comparative analysis of modern machine learning models for retail sales forecasting
Luka Hobor
; Faculty of Electrical Engineering and Computing, University of Zagreb, Zagreb, Croatia
*
Mario Brčić
; Faculty of Electrical Engineering and Computing, University of Zagreb, Zagreb, Croatia
Lidija Polutnik
; Babson College, Babson Park, Massachusetts, United States
Ante Kapetanović
; mStart Plus d.o.o., Zagreb, Croatia
* Corresponding author.
Abstract
Accurate demand forecasting is critical for brick-and-mortar retailers to optimize inventory management and minimize costs. This study evaluates statistical baselines, tree-based ensembles (XGBoost and LightGBM), and deep learning architectures (N-BEATS, N-HiTS, and the Temporal Fusion Transformer) on retail sales data characterized by intermittent demand, substantial missingness, and frequent product turnover. Models are compared across four configurations varying by aggregation level and imputation strategy, using evaluation protocols that reflect typical deployment patterns for each model class. Localized tree-based methods achieve superior performance, with XGBoost attaining the lowest RMSE of 4.833. While SAITS-based imputation improved neural network performance in aggregated settings, these models remained inferior to ensemble methods. The results suggest that, under the studied constraints, model selection should prioritize alignment with problem characteristics over architectural sophistication.
Keywords
gradient-boosted decision trees; neural networks; predictive analytics; retail sales forecasting; time-series analysis
Hrčak ID:
347148
URI
Publication date:
13.5.2026.
Visits: 0 *