Comparative analysis of machine learning models for retail demand forecasting considering promotional activity and seasonality

Nataliya Boyko*, Arsen Dolichnyi

nataliya.i.boyko@gmail.com

Received 24.09.2025, Revised 10.03.2026, Accepted 28.04.2026, Published 29.05.2026
Abstract

The relevance of the study is driven by the complexity of modelling consumer behaviour in an unstable market environment, where forecast accuracy directly impacts inventory optimisation and the reduction of logistical costs. The aim of this study was to develop and substantiate an AI-based intelligent system for demand forecasting and procurement optimisation in retail, ensuring consistent product availability while simultaneously reducing excess inventory and waste. The research methodology was based on a rigorous time-based data split, which allows for evaluating the models’ generalisation ability in a realistic “train on the past, test on the future” scenario. To establish a robust point of reference, a hierarchy of baseline benchmarks was developed, including naive forecasts and models with weekly lags. The study conducted a comparative analysis of Ridge linear regression, the Random Forest ensemble method, and gradient-boosted decision trees (XGBoost, LightGBM). Experimental results confirmed a significant advantage of non-linear models over traditional linear approaches, which fail to adequately reproduce threshold effects and complex interactions between promotions and seasonality. It was established that the LightGBM model demonstrates the best stability and accuracy metrics, ensuring minimal error on the validation set while maintaining resistance to overfitting. Specifically, it was found that XGBoost’s high efficiency on training data is often a result of over-adaptation to noise, making LightGBM more suitable for practical business applications. The practical significance of the findings lies in the potential for automating procurement processes, ensuring balanced inventory levels and increasing the economic efficiency of retail networks

Keywords:
predictive analytics; feature engineering; gradient boosting; time series; mean absolute error; root mean squared error; business analytics

Retrieved from Volume 24, No. 1, 2026

Pages 35–53

References Suggested citation
  1. Agga, A., Abbou, A., Labbadi, M., El Houm, Y., & Ou Ali, I.H. (2022). CNN-LSTM: An efficient hybrid deep learning architecture for predicting short-term photovoltaic power production. Electric Power Systems Research, 208, article number 107908. doi: 10.1016/j.epsr.2022.107908.
  2. Bülbül, K., & Noyan, N. (2021). Multi-stage stochastic programming models for provisioning cloud computing resources. European Journal of Operational Research, 288(3), 834-846. doi: 10.1016/j.ejor.2020.06.027.
  3. Corporación Favorita grocery sales forecasting. (2017). Retrieved from https://www.kaggle.com/c/favorita-grocery-sales-forecasting.
  4. Djaafari, A., Ibrahim, A., Bailek, N., Bouchouicha, K., Hassan, M.A., Kuriqi, A., Al-Ansari, N., & El-kenawy, El-S. (2022). Hourly predictions of direct normal irradiation using an innovative hybrid LSTM model for concentrating solar power projects in hyper-arid regions. Energy Reports, 8, 15548-15562. doi: 10.1016/j.egyr.2022.10.402.
  5. Fortin, A.-P., Simonato, J.-G., & Dionne, G. (2023). Forecasting expected shortfall: Should we use a multivariate model for stock market factors? International Journal of Forecasting, 39(1), 314-331. doi: 10.1016/j.ijforecast.2021.11.010.
  6. Guo, D., Ye, L., Yan, G., & Min, H. (2022). CCD-BSM: Composite-curve-dilation brush stroke model for robotic Chinese calligraphy. Applied Intelligence, 53, 14269-14283. doi: 10.1007/s10489-022-04210-y.
  7. Haque, M.S., Amin, M.S., & Miah, J. (2023). Retail demand forecasting: A comparative study for multivariate time series. ArXiv. doi: 10.48550/arXiv.2308.11939.
  8. Ketelsen, M., Janssen, M., & Hamm, U. (2020). Consumers’ response to environmentally-friendly food packaging – a systematic review. Journal of Cleaner Production, 254, article number 120123. doi: 10.1016/j.jclepro.2020.120123.
  9. Kim, D., & Jang, L.-Ch. (2023). Economic preference for semiconductor trade deals using similarity measures defined by Choquet integrals. Computational and Applied Mathematics, 42, article number 224. doi: 10.1007/s40314-023-02365-z.
  10. Moret, S., Babonneau, F., Bierlaire, M., Maréchal, F. (2020). Decision support for strategic energy planning: A robust optimisation framework. European Journal of Operational Research, 280(2), 539-554. doi: 10.1016/j.ejor.2019.06.015.
  11. Rashid, M., Kamruzzaman, J., & Imam, T. (2022). A tree-based stacking ensemble technique with feature selection for network intrusion detection. Applied Intelligence, 52, 9768-9781. doi: 10.1007/s10489-021-02968-1.
  12. Slama, S.B., & Mahmoud, M. (2023). A deep learning model for intelligent home energy management system using renewable energy. Engineering Applications of Artificial Intelligence, 123(B), article number 106388. doi: 10.1016/j.engappai.2023.106388.
  13. Solodkyi, V., & Polichuk, Yu. (2023). Implementation of artificial intelligence in Ukrainian banks: Prospects and limitations. Finance, Accounting and Taxation, 82(2), 119-127. doi: 10.33271/ebdut/82.119.
  14. Tama, B.A., & Lim, S. (2021). Ensemble learning for intrusion detection systems: A systematic mapping study and cross-benchmark evaluation. Computer Science Review, 39, article number 100357. doi: 10.1016/j.cosrev.2020.100357.
  15. Thudumu, S., Branch, P., Jin, J., & Singh, J. (2020). A comprehensive survey of anomaly detection techniques for high dimensional big data. Journal of Big Data, 7, article number 42. doi: 10.1186/s40537-020-00320-x.
  16. Uninets, I. (2021). Subjective disposition of smart economy. Economics and Education, 6(2), 12-16. doi: 10.30525/2500-946X/2021-2-2.
  17. Xie, H., Hao, C., Li, J., Li, M., Luo, P., & Zhu, J. (2022). Anomaly detection for time series data based on multi-granularity neighbor residual network. International Journal of Cognitive Computing in Engineering, 3, 180-187. doi: 10.1016/j.ijcce.2022.10.001.
  18. Xu, X., Fang, Z., Qi, L., Zhang, X., He, Q., & Zhou, X. (2021). TripRes: Traffic flow prediction driven resource reservation for multimedia IoV with edge computing. ACM Transactions on Multimedia Computing, Communications, and Applications, 17(2), article number 41. doi: 10.1145/3401979.
  19. Yan, Yu., Fang, Sh., Gao, M., & Chen, Y. (2024). MUS model: A deep learning-based architecture for IoT intrusion detection. Computers, Materials & Continua, 80(1), 875-896. doi: 10.32604/cmc.2024.051685.
  20. Yang, Y., Zhang, Ch., Zhao, Q., & Zhang, Yu. (2024). A sustainability evaluation framework for the urban energy Internet using the Fermatean fuzzy Aczel-Alsina hybrid MCDM method. Expert Systems with Applications, 238(D), article number 122115. doi: 10.1016/j.eswa.2023.122115.
  21. Yin, J., Shi1, Y., Deng, W., Yin, Ch., Wang, T., Song, Y., Li, T., & Li, Y. (2023). Internet of Things intrusion detection system based on convolutional neural network. Computers, Materials & Continua, 75(1), 2119-2135. doi: 10.32604/cmc.2023.035077.
Boyko, N., & Dolichnyi, A. (2026). Comparative analysis of machine learning models for retail demand forecasting considering promotional activity and seasonality. Journal of Kryvyi Rih National University, 24(1), 35-53. https://doi.org/10.31721/2306-5451-2026-1-24-35-53
en