Time Series-Based Demand Forecasting: A Comparative Analysis of Holt-Winters, SARIMA, and Prophet Models on Retail Inventory Data

Alican Doğan

Makale detayları

Time Series-Based Demand Forecasting: A Comparative Analysis of Holt-Winters, SARIMA, and Prophet Models on Retail Inventory Data

Yazarlar:

Alican Doğan

Yayıncı:

Istanbul Ticaret University

Ana Başlıklar:

İş Analitiği

Dergi:

İstanbul Ticaret Üniversitesi Sosyal Bilimler Dergisi Vol.24 Issue 53, pp.641-668

Belge Tipi:

Makale

Yayın Tarihi:

9/2025

Belge Dili:

İngilizce

Document ID (Org No.):

2001359385

pISSN:

1303-5495

Anahtar kelimeler:

Özet:

 Accurate inventory management is a critical component of operational efficiency in the retail sector, directly influencing customer satisfaction and logistics cost optimization. Demand forecasting plays a pivotal role in this process by enabling businesses to anticipate future product needs and make data-driven decisions. This study aims to evaluate and compare the performance of classical time series models for demand forecasting using a rich and structured dataset. The analysis is based on the Retail Store Inventory Forecasting Dataset, a synthetic but realistic collection comprising over 73,000 daily sales records across multiple products and retail locations. Three prominent time series forecasting methods were selected for this research: the Holt-Winters Exponential Smoothing Model, the Seasonal ARIMA SARIMA model, and Prophet. These models were implemented to predict daily product demand and evaluated on a test set using Mean Absolute Error MAE, Root Mean Squared Error RMSE, and Mean Absolute Percentage Error MAPE as performance metrics. Among the three, the Holt-Winters model delivered the best forecasting performance with the lowest error values. The findings reveal that classical time series models remain powerful tools for retail forecasting tasks, especially when capturing patterns driven by seasonality and trend. Furthermore, this study demonstrates that robust forecasting techniques can substantially support inventory optimization efforts, helping to mitigate common challenges such as overstocking and stockouts. By highlighting the practical value of interpretable and well-established models, this research provides a foundational perspective for integrating time series forecasting into business intelligence and decision support systems.

 Accurate inventory management is a critical component of operational efficiency in the retail sector, directly influencing customer satisfaction and logistics cost optimization. Demand forecasting plays a pivotal role in this process by enabling businesses to anticipate future product needs and make data-driven decisions. This study aims to evaluate and compare the performance of classical time series models for demand forecasting using a rich and structured dataset. The analysis is based on the Retail Store Inventory Forecasting Dataset, a synthetic but realistic collection comprising over 73,000 daily sales records across multiple products and retail locations. Three prominent time series forecasting methods were selected for this research: the Holt-Winters Exponential Smoothing Model, the Seasonal ARIMA SARIMA model, and Prophet. These models were implemented to predict daily product demand and evaluated on a test set using Mean Absolute Error MAE, Root Mean Squared Error RMSE, and Mean Absolute Percentage Error MAPE as performance metrics. Among the three, the Holt-Winters model delivered the best forecasting performance with the lowest error values. The findings reveal that classical time series models remain powerful tools for retail forecasting tasks, especially when capturing patterns driven by seasonality and trend. Furthermore, this study demonstrates that robust forecasting techniques can substantially support inventory optimization efforts, helping to mitigate common challenges such as overstocking and stockouts. By highlighting the practical value of interpretable and well-established models, this research provides a foundational perspective for integrating time series forecasting into business intelligence and decision support systems.

 Accurate inventory management is a critical component of operational efficiency in the retail sector, directly influencing customer satisfaction and logistics cost optimization. Demand forecasting plays a pivotal role in this process by enabling businesses to anticipate future product needs and make data-driven decisions. This study aims to evaluate and compare the performance of classical time series models for demand forecasting using a rich and structured dataset. The analysis is based on the Retail Store Inventory Forecasting Dataset, a synthetic but realistic collection comprising over 73,000 daily sales records across multiple products and retail locations. Three prominent time series forecasting methods were selected for this research: the Holt-Winters Exponential Smoothing Model, the Seasonal ARIMA SARIMA model, and Prophet. These models were implemented to predict daily product demand and evaluated on a test set using Mean Absolute Error MAE, Root Mean Squared Error RMSE, and Mean Absolute Percentage Error MAPE as performance metrics. Among the three, the Holt-Winters model delivered the best forecasting performance with the lowest error values. The findings reveal that classical time series models remain powerful tools for retail forecasting tasks, especially when capturing patterns driven by seasonality and trend. Furthermore, this study demonstrates that robust forecasting techniques can substantially support inventory optimization efforts, helping to mitigate common challenges such as overstocking and stockouts. By highlighting the practical value of interpretable and well-established models, this research provides a foundational perspective for integrating time series forecasting into business intelligence and decision support systems.

 Perakende sektöründe stok seviyelerinin doğru bir şekilde yönetilmesi, müşteri memnuniyetinin artırılması ve lojistik maliyetlerin azaltılması açısından büyük önem taşımaktadır. Bu doğrultuda, talep tahmini modellerinin doğruluğu, işletmelerin karar destek sistemlerinde kritik bir rol oynamaktadır. Bu çalışmada, perakende mağazalarında ürün talebini zaman serisi analiz yöntemleriyle tahmin etmek amacıyla klasik modellere dayalı bir karşılaştırma yapılmıştır. Analizlerde, 73.000den fazla günlük satış kaydı içeren Retail Store Inventory Forecasting Dataset adlı sentetik ancak gerçekçi bir veri seti kullanılmıştır. Çalışma kapsamında üç farklı zaman serisi modeline odaklanılmıştır: Holt-Winters Üstel Düzeltme Modeli, Mevsimsel ARIMA SARIMA ve Prophet. Bu modeller, günlük ürün satışlarını tahmin etmek üzere uygulanmış ve performansları Ortalama Mutlak Hata MAE, Kök Ortalama Kare Hata RMSE ve Ortalama Mutlak Yüzde Hata MAPE metrikleri ile test verisi üzerinde değerlendirilmiştir. Model çıktıları karşılaştırıldığında, Holt-Winters modeli, en düşük hata oranlarını vererek en başarılı tahmin performansını göstermiştir. Elde edilen bulgular, özellikle mevsimsellik ve trend gibi yapısal bileşenlerin öne çıktığı perakende satış verilerinde, klasik zaman serisi modellerinin hala güçlü ve etkili araçlar olduğunu ortaya koymaktadır. Bu çalışma, işletmelere daha doğru talep tahminleri ile stok fazlası ve stok yetersizliği gibi sorunları azaltma fırsatı sunmakta ve zaman serisi tabanlı yaklaşımların karar destek sistemlerinde nasıl kullanılabileceğine dair değerli bir çerçeve önermektedir.

Makale detayları

Time Series-Based Demand Forecasting: A Comparative Analysis of Holt-Winters, SARIMA, and Prophet Models on Retail Inventory Data

Yazarlar:

Alican Doğan

Yayıncı:

Istanbul Ticaret University

Ana Başlıklar:

İş Analitiği

Dergi:

İstanbul Ticaret Üniversitesi Sosyal Bilimler Dergisi Vol.24 Issue 53, pp.641-668

Belge Tipi:

Makale

Yayın Tarihi:

9/2025

Belge Dili:

İngilizce

Document ID (Org No.):

2001359385

pISSN:

1303-5495

Time series analysis

Inventory management

Demand forecastingZaman serisi analizi

Stok yönetimi

Talep tahmini

Sayı

Makale detayları

Time Series-Based Demand Forecasting: A Comparative Analysis of Holt-Winters, SARIMA, and Prophet Models on Retail Inventory Data

Yazarlar: Alican Doğan

Yayıncı: Istanbul Ticaret University

Ana Başlıklar: İş Analitiği

Dergi: İstanbul Ticaret Üniversitesi Sosyal Bilimler Dergisi Vol.24 Issue 53, pp.641-668

Belge Tipi: Makale

Yayın Tarihi: 9/2025

Belge Dili: İngilizce

Document ID (Org No.): 2001359385

pISSN: 1303-5495

Time series analysis

Inventory management

Demand forecastingZaman serisi analizi

Stok yönetimi

Talep tahmini

Sayı

Yazarlar:

Alican Doğan

Yayıncı:

Istanbul Ticaret University

Ana Başlıklar:

İş Analitiği

Dergi:

İstanbul Ticaret Üniversitesi Sosyal Bilimler Dergisi Vol.24 Issue 53, pp.641-668

Belge Tipi:

Makale

Yayın Tarihi:

9/2025

Belge Dili:

İngilizce

Document ID (Org No.):

2001359385

pISSN:

1303-5495