A new method of diesel fuel brands identification: SMOTE oversampling combined with XGBoost ensemble learning.

• A new model combined with NIR spectroscopy is proposed to identify diesel oil brands. • XGBoost ensemble learning successfully applied in NIR spectroscopy field. • SMOTE oversampling technology has advantages in balancing unbalanced samples. • 5 models are compared for diesel brands identification. • The proposed model has the highest recognition rate. • The proposed model can improve the recognition accuracy of rare samples. Using proper diesel brand is the key to ensure the normal operation of diesel engine. It is even more important to identify the brands of diesel oil effectively. This paper presented a new model of near infrared spectroscopy (NIRS) identification of diesel oil brands that combined Tree-based feature selection, Synthetic Minority Oversampling Technique (SMOTE) and Extreme Gradient Boosting (XGBoost) ensemble learning in order to achieve the goal of high accuracy and rapidity. To further demonstrate the practical effect of the proposed ensemble method, it was compared with a single decision tree (DT) classifier based on classification and regression tree (CART) algorithm. As a result, the recognition rate of Tree-SMOTE-XGBoost model proposed in this paper was 19.33% higher than that of XGBoost model, and 9.25% higher than that of Tree-SMOTE-DT model. More importantly, it can ensure the accuracy of each class under the premise of serious imbalance of classes. The proposed method saves manpower and material resources, and provides a new alternative approach for diesel brands identification. [ABSTRACT FROM AUTHOR]