Machine learning in supply chain: prediction of real-time e-order arrivals using ANFIS

Accurate demand forecasting throughout the multi-channel supply chain (SC) enhances the managers' decision-making capability in operational, tactical, and strategic aspects. However, the problem is that earlier publications about the real-time prediction of e-commerce order arrivals in the SC show some inadequacies. According to a systematic review from Tsolaki (ICT Express, 2022. ) who integrate logistics and machine learning (ML) methods in the past ten years, there are very few studies that focus on arrival time prediction like this study does, and none of them uses an adaptive neuro-fuzzy inference system (ANFIS) framework to predict e-order arrivals. Besides, (Policarpo in Comput Sci Rev 41:100414, 2021) review the existing publications that integrate e-commerce and ML techniques in the past five years; they reveal that previous studies pay heavier attentions to e-commerce initiative goals such as purchase and repurchase predictions, and none of them focuses on predicting e-order arrivals like this study does. Previous scholars investigate SC orders and prediction issues in a broader space, while this study attempts to predict hour-to-hour, actual-time order arrivals. Thus, this study presents a new data-empowered forecasting method to fill these research gaps. The motivation of this study is to build a method for predicting real-time e-orders arrivals in distribution hubs, enabling third-party logistics providers to handle the hourly-based e-order arrival rates more efficiently. This study tries to find the solution for the problem by developing a new ML forecasting method by integrating time-series data features and ANFIS, which has been proven to significantly reduce the issues' computational complexity. This study creates a four-phase operation model to enable managers to adopt the suggested framework, and develops a systematized forecasting model to cross-confirm the framework's outcomes. This study employs a descriptive case study and shows a satisfactory degree of precision of the suggested ML method in predicting the actual e-order arrivals in three e-retailers at three-hour cycle times. The findings reveal that the real-time forecasting is significant to boost the values of e-order arrivals in every day business operations. The novelty of this study lies on its novel contribution and purpose to build a method for predicting real-time e-orders arrivals in distribution hubs, enabling third-party logistics providers to handle the hourly-based e-order arrival rates more efficiently; and to develop a new ML forecasting method by integrating ANFIS and time-series data features.