Early prediction of whether a product will go to backorder or not is necessary for optimal management of inventory that can reduce the losses in sales, establish a good relationship between the supplier and customer and maximize the revenues. In this study, we have investigated the performance and effectiveness of tree based machine learning algorithms to predict the backorder of a product. The research methodology consists of preprocessing of data, feature selection using statistical hypothesis test, imbalanced learning using the random undersampling method and performance evaluating and comparing of four tree based machine learning algorithms including decision tree, random forest, adaptive boosting and gradient boosting in terms of accuracy, precision, recall, f1-score, area under the receiver operating characteristic curve and area under the precision and recall curve. Three main findings of this study are (1) random forest model without feature selection and with random undersampling method achieved the highest performance in terms of all performance measure metrics, (2) feature selection cannot contribute to the performance enhancement of the tree based classifiers, and (3) random undersampling method significantly improves performance of tree based classifiers in product backorder prediction.