Classification of Customer Orders in The Internal Section of Supply Chain Management Using Machine Learning

Abstract

Customizing orders through the marketplace results in a very large number of product variants that must be made by manufacturers. Product customization that is too far from product standards can cause losses. So far, the manufacturer knows the loss when the order has been received and paid for by the consumer. The marketplace application cannot classify the types of orders that can or cannot be produced. Orders that have been received cannot be canceled by the manufacturer because it can lower the rating and credibility of the manufacturer. The use of machine learning in marketplace applications with random forest algorithms can classify order data, whether they can or cannot be produced. The results of the study prove that the rendom forest model made for order classification has accuracy=100%, sensitivity=100%, and specificity=100% for the dataset of batik shirt orders from consumers. Predictions are made based on order specifications, such as quantity, gender, size, collar type, cloth material, and sleeve type. The accuracy of the prediction results is also achieved by using the value of the number of trees (ntree) 50 with mtry 2. The dataset is in the form of order data as many as 3039 records taken within 6 weeks.