Machine learning-based sales forecasting during crises: Evidence from a Turkish women's clothing retailer.

Area of Science:

• Retail Analytics
• Supply Chain Management
• Machine Learning Applications

Background:

• Natural disasters and pandemics significantly disrupt retail supply chains, causing shortages and financial strain.
• The COVID-19 pandemic underscored the need for retailers to adapt sales forecasts to volatile conditions.
• Regular assessment of sales volumes and consumer behavior is crucial for retail crisis preparedness.

Purpose of the Study:

• To explore strategies for adapting retail sales forecasts during crises.
• To analyze consumer behavior shifts and sales impacts across diverse product categories.
• To evaluate the effectiveness of various machine learning methods in retail forecasting.

Main Methods:

• Utilized machine learning (ML) algorithms to analyze sales data and consumer behavior.
• Examined product categories including apparel, accessories, and footwear.
• Assessed performance of Gradient Boosting, CatBoost, Multi-Layer Perceptron (MLP), LightGBM, and XGBoost.

Main Results:

• Gradient Boosting and CatBoost excelled in high-sales-change categories.
• MLP demonstrated effectiveness in low-volume categories like accessories and footwear.
• MLP, LightGBM, and XGBoost were successful in medium-volume categories such as outerwear and underwear.

Conclusions:

• Machine learning models effectively adapt sales forecasts to crisis conditions.
• Findings provide a practical approach to enhance retail resilience against disruptions.
• The study offers an effective method for adapting sales forecasting to evolving consumer behaviors during crises.