A machine learning based approach for backorders prediction: A study of NPT e-commerce company

Abstract

NPT is a corporation that specializes in E-commerce cooperation. Their primary lines of business include retail sales and operations, digital marketing, customer service, warehousing, and fulfillment. The company profit is generated by addressing customers' needs for beauty products such as cosmetics, skincare products, etc. Because the cost of holding stock grows exponentially with increased availability which will never reach 100%. This effectively means that there will be times when an order is placed for an item that is currently out of stock, resulting in a backorder. While not all backorders may be totally avoided, anticipating them allows for proactive efforts to be made, potentially lowering lead times and costs. To address the problem of backorder prediction, this study compared various machine learning models for binary classification. There are three machine learning models (Light Gradient Boosting machine, Random Forest, and CatBoost) that are evaluated based on their ability to predict backorders. Additionally, one data transformation technique is used (Normalization) comes along with AUC score and performance metrics comparison. Among these models, RF performed the best after scrutinizing the sensitivity. The features’ importance score analysis highlighted the significance of inventory stock, product delivery volume, imminent demand (sales), and accurate future demand prediction in correctly predicting backorders.