Softening Bullwhip Effect by Improving Sales Forecast Accuracy Using Machine Learning

The bullwhip effect is recognized as one of the main inefficiencies in supply chain organizations because it has significant consequences for cash flow and company sustainability. One of the causes of the bullwhip effect is the low forecast accuracy of the company's forecasting system. The inaccuracies in demand forecasting are attributed to distortions in information flow within the supply chain. The objective of this study is to quantify the Bullwhip effect value (BWE), find the predicting model that is the most accurate, and Identify a suitable model that can be applied to soften the bullwhip effect. In this work, we will identified the bullwhip effect value by generating from sales dataset manufacturing company. Using machine learning we employ various forecasting methods: ARMA, ARIMA, Adaboost, ANN, KNN, Random Forest, SVM, VAR and Xgboost to resolve the bullwhip effect. According to the study's findings, Xgboost and Adaboost outperform other forecasting methods in terms of accuracy, with lower Mean Absolute Percentage Error (MAPE) (0.37 and 0.24) and Root Mean Squared Error (RMSE) (14,438 and 41,228) as well as higher R2 (0.99 and 0.95). The most significant reduction in BWE was obtained by using Xgboost to compare the real BWE value of 8,12 to 1,23. It is possible to conclude that using the most accurate prediction model will reduce the bullwhip effect.