Comparative Analysis of Modern Machine Learning Models for Retail Sales Forecasting

TL;DR

This paper compares various machine learning models for retail sales forecasting, finding that ensemble methods like XGBoost outperform deep learning architectures in accuracy, especially under real-world data challenges.

Contribution

It provides a comprehensive evaluation of statistical, tree-based, and deep learning models on retail demand data with practical complexities, highlighting the importance of model selection based on problem characteristics.

Findings

XGBoost achieved the lowest RMSE of 4.833.

Tree-based models outperformed neural networks in this context.

Imputation strategies improved neural network performance but did not surpass ensemble methods.

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.