Accelerate demand forecasting by hybridizing CatBoost with the dingo optimization algorithm to support supply chain conceptual framework precisely

Abstract

Supply chains (SCs) serve many sectors that are, in turn, affected by e-commerce which rely on the make-to-order (MTO) system to avoid a risk in following the make-to-stoke (MTS) policy due to poor forecasting demand, which will be difficult if the products have short shelf life (e.g., refrigeration foodstuffs). The weak forecasting negatively impacts SC sectors such as production, inventory tracking, circular economy, market demands, transportation and distribution, and procurement. The forecasting obstacles are in e-commerce data types that are massive, imbalanced, and chaotic. Using machine learning (ML) algorithms to solve the problem works well because they quickly classify things, which makes accurate forecasting possible. However, it was found that the accuracy of ML algorithms varies depending on the SC data sectors. Therefore, the presented conceptual framework discusses the relations among ML algorithms, the most related sectors, and the effective scope of tackling their data, which enables the companies to guarantee continuity and competitiveness by reducing shortages and return costs. The data supplied show the e-commerce sales that were made at 47 different online stores in Egypt and the KSA during 413 days. The article proposes a novel mechanism that hybridizes the CatBoost algorithm with Dingo Optimization (Cat-DO), to obtain precise forecasting. The Cat-DO has been compared with other six ML algorithms to check its superiority over autoregressive integrated moving average (ARIMA), long short-term memory (LSTM), deep neural network (DNN), categorical data boost (CatBoost), support vector machine (SVM), and LSTM-CatBoost by 0.52, 0.73, 1.43, 8.27, 15.94, and 13.12%, respectively. Transportation costs were reduced by 6.67%.