Article Details

Abstract

To solve the problems of sparse fault samples and class imbalance in industrial equipment data, a predictive maintenance model based on SMOTE-XGBoost is developed. Key features such as temperature, rotational speed, torque and wear are extracted from a dataset in the UCI Machine Learning Repository, and SMOTE oversampling is adopted to balance the training samples, thus effectively suppressing the prediction bias of the model. In comparisons with multiple mainstream models under the 5-fold cross-validation framework, the results show that SMOTE-XGBoost achieves the optimal performance, with an accuracy of 0.986, a precision of 0.88, a recall of 0.79 and an F1-score of 0.83. This scheme significantly improves the identification accuracy of minority-class faults under complex working conditions and provides solid data-driven support for practical engineering scenarios.

Keywords

fault identification; predictive maintenance; class imbalance; XGBoost

References

1. [1] Bromberger D，Júnior A N，Garcia V，et al. Interpretable Machine Learning in Industrial Water Pump Failure Prediction: The Role of Shapley Additive Explanations Values and Data Sampling Techniques [J/OL].Engineering Applications of Artificial Intelligence，2026，165：113558.https://doi.org/10.1016/j.engappai.2025.113558.
2. [2] Raj R，Kolhe K，Gutte V，et al. IoT-Based Predictive Maintenance Using AI/ML: A Systematic Review [J/OL].AIP Conference Proceedings，2024，3028：020035.https://doi.org/10.1063/5.0212418.
3. [3] Harsh T，Kumar T，Mohanty A，et al. Predictive Maintenance of Industrial Machines Using ML and IoT [C]//2024 3rd International Conference on Artificial Intelligence for Internet of Things (AIIoT)，2024：1-6.
4. [4] Wang Yujing, Ye Baihong, Kang Shouqiang, etc Fault diagnosis method for harmonic reducers based on federated learning under multi-source unbalanced data [J]. Journal of Instrumentation, 2025, 46 (6): 317-329
5. [5] Zou X，Dai J. Feature Selection Based on Rough Diversity Entropy [J/OL].Pattern Recognition，2026，170：112032.https://doi.org/10.1016/j.patcog.2025.112032.
6. [6] UCI Machine Learning Repository. AI4I 2020 Predictive Maintenance Dataset [DB/OL].[2025-10-16].https://archive.ics.uci.edu/dataset/601/ai4i+2020+predictive+maintenance+dataset.
7. [7] Prasad M，Arepalli M K，Varma P V N N，et al. Probrank: A Feature Probability Estimation-Based Framework for Feature Selection and Ranking [J].Evolutionary Intelligence，2026，19（1）.https://doi.org/10.1007/s12065-025-01118-7.
8. [8] Wang X，Zhou S，Hou B，et al. A High-Dimensional Feature Selection Method Based on Feature Interaction Clustering and Integer-Encoded TLBO [J].Computational and Applied Mathematics，2026，45（1）：1-28.
9. [9] Elor Y，Averbuch-Elor H. To SMOTE, or Not to SMOTE? [PP/OL].arXiv.（2022-05-11）.https://arxiv.org/abs/2201.08528.
10. [10] Chen Yafeng, Zhang Zhidi, Xue Yan, etc Adversarial verification method based on regularized 5 × 2 cross validation voting [J]. Computer Technology and Development, 2026, 36 (8): 78-86, 95
11. [11] Ma Heng, Zhang Shilong, High Tech A temperature prediction model for excavation face based on ensemble learning reinforced BPNN [J]. Industrial and Mining Automation, 2025, 51 (8): 88-94158
12. [12] Wang Yihan, Wang Zongren, Li Youli, etc Quality Control Method for Solar Greenhouse Environmental Sensing Data Based on LSTM XGBoost [J]. Journal of Agricultural Engineering, 2026, 42 (6): 254-262
13. [13] Leng Qiangkui, Sun Xuezi, Meng Xiangfu A boundary sample synthesis oversampling method based on KNN and random affine [J]. Journal of Intelligent Systems, 2025, 20 (2): 329-343

About the authors

Zuo Hangxu (1998-), male, Han nationality, from Luzhou, Sichuan, lecturer, master's degree, research direction: machine learning, deep learning and their applications in healthcare and industry; Yuan Shichang (1996-), male, Han ethnicity, from Xuchang, Henan Province, China, lecturer, master's degree, research direction: data mining, network security and intelligent information processing technology, feature modeling and security analysis methods in complex data environments.