مدیریت تولید و عملیات (Mar 2023)
Proposing an integrated approach for omnichannel demand forecasting using machine learning-time series clustering with dynamic time warping algorithm and artificial neural networks
Abstract
Purpose: The increasing complexity of omnichannel retailing has necessitated retailers to redesign processes and forecasting methods and accept new approaches based on machine learning and artificial intelligence. Improving the accuracy of demand forecasting and managing customer needs from different channels due to reducing demand uncertainty are the most important challenges in omnichannel retailing that retailers should deal with. A better understanding of consumer behaviour patterns leads to more accurate demand forecasting, which in turn helps gain insight into transportation flows, improves distribution management, and enables better planning and execution of supply chain operations. This study aims to reduce the uncertainty of demand in omnichannel retailing by improving the accuracy of demand forecasting by considering customers buying behaviour through using machine learning methods. Design/methodology/approach: In this study to forecast future sales based on customers buying behaviour, a cosmetics retailer’s historical data on the monthly sales from February 2020 to June 2022 is used. The ID of eight products has been selected to analyze the performance of proposed methods and the method that the company applied to forecast demand. Clustering has been implemented using the dynamic time-warping algorithm due to the unequal length of the products’ time series. Initially, the nonlinear autoregressive neural network (NAR) has been applied to the time series in each cluster and later, the nonlinear autoregressive neural network with exogenous input (NARX) has been applied to the time series. The performance of the methods has been evaluated by testing R-squared and all R-squared coefficients and root mean square error (RMSE) to analyze the accuracy measure. Findings: The forecasting methods comparison, moving average (MA), the nonlinear autoregressive neural network (NAR), and the nonlinear autoregressive neural network with exogenous input (NARX) concerning testing R-squared coefficient, and also all R-squared and RMSE indicated that the nonlinear autoregressive neural network with exogenous input presented a good performance for all the products, so it confirmed that the application of the clustering to identification customers buying behaviour through the sales history of the products, integrated with artificial neural networks, to conduct demand forecasting, could be considered a good method for forecasting demand of omnichannel retailing supply chain products. Practical implications: The proposed method of this study leads to uncertainty reduction in omnichannel retailing by understanding the buying behaviour of customers, identifying patterns and using its analysis in the processes and operations, and its integration with machine learning methods improves distribution management and provides better planning and implementation of supply chain operations. Managers can use the proposed method to accurately predict complex demand patterns in the retailing industry. Using business data in demand planning provides an extra advantage to managers to include important variables based on their judgments. Social implications: Knowing the factors affecting the sale of a specific category of a product helps to effectively design promotions, advertising campaigns, the optimal combination of category displays and optimization of shelf space in retail stores. Also, accurate demand forecasts lead to better ordering policies, thus minimizing the cost of inventory management and optimal distribution and logistics planning to meet future demand. Originality/value: The proposed method presents a predictive approach for an omnichannel retailing supply chain that leads to uncertainty reduction in omnichannel retailing by understanding the buying behaviour of customers, identifying patterns, and using its analysis in the processes and operations and its integration with machine learning methods to improve distribution management, and provides better planning and implementation of supply chain operations.
Keywords
