Data-driven rated power prediction of diesel engines using improved multi-class imbalanced learning method
Liangxun Guo, Zilong Zhuang, Yan‐Ning Sun, Wei Qin · Procedia Manufacturing · 2020
Rated power is one of the most important quality indicators of diesel engines. The effective prediction of the rated power classes is of great significance in controlling the quality of diesel engines. However, the number of samples varies greatly in different classes. Creating an effective machine learning model to predict the rated power classes is difficult due to the imbalanced dataset. This paper develops an improved multi-class imbalanced learning method which integrates dynamic oversampling method with multi-class AdaBoost for producing high predictive accuracy over both the majority classes and the minority classes of the diesel engine dataset. Computational experiments have been performed on four UCI standard datasets and one diesel engine’s rated power dataset, and the experimental results indicate that the proposed algorithm is superior to the compared methods, including resampling methods and boosting-type methods.