A Comparative Analysis Of Machine Learning Algorithms To Predict Backorder In Supply Chain Management

Abstract

In supply chain inventory management, the backorder of products is a common occurrence. A backorder can only be successful if the supplier can be relied upon to provide the out-of-stock product when required or if the consumer is prepared to wait for the product without becoming impatient. So, to avoid uncertain circumstances and lessen the loss, it is important to know which goods are more likely to be back-ordered ahead of time so that appropriate measures can be taken. This work presents a backorder predictive model based on an imbalanced historical dataset obtained from an online source. The data has undergone sampling techniques, various Machine Learning algorithms, and evaluation metrics. After Random Undersampling (RUS), Random Forest (RF) achieved the highest accuracy of 99.328% while an accuracy of 98.917% was obtained from Extreme Gradient Boosting (XGBoost) model, which was highest after applying Synthetic Minority Oversampling Technique (SMOTE). XGBoost has outperformed other models with highest ROC-AUC score of 94.55% and 95.93% for oversampled and undersampled data respectively. In this paper, we made a visualization of the attributes of the dataset, then applied and analyzed Machine Learning algorithms to predict the backorder.