Retail Products Price Forecasting with Empirical Mode Decomposition and Auto Regressive Integrated Moving Average Model Using Web-Scraped Price Microdata

This study presents a cutting-edge approach to price forecasting for an online retail business in Turkey, utilizing a hybrid model that combines Empirical Mode Decomposition (EMD) with Auto Regressive Integrated Moving Average (ARIMA) models. A 900-day dataset, scraped from the website, underpins this analysis. A battery of fourteen metrics is employed to evaluate the forecasting performance, culminating in a statistically significant confirmation of the hybrid model's superiority over the standalone ARIMA model, as established by the Wilcoxon signed-rank test. In addition to this performance validation, our investigation unveils an intriguing association between category standard deviations and forecasting accuracy, with lower standard deviations correlating with higher forecasting performance. While acknowledging the study's limitations related to data collection constraints, this research bears wider significance for the entire supply chain, offering strategic insights for retailers and the potential for more detailed analysis with larger datasets. Moreover, it lays the groundwork for future studies involving dynamic ARIMA parameter determination, advanced EMD variants, and machine learning integration, enhancing its applicability to various time series contexts. The results are compared with machine learning algorithms, namely Neural Networks, Support Vector Regression, Regression Tree, Gaussian Process Regression, and the Generalized Additive Model.