Abstract: This paper introduces several traditional Support Vector Machine (SVM) multi-class classification methods and analyzes their existing problems and shortcomings. A binary decision tree-based SVM multi-class classification method is proposed, which is simple, intuitive, and requires fewer training samples. By applying this method to fault diagnosis of hoist operation status, the results show that it achieves a higher diagnostic accuracy than traditional multi-class SVM methods and BP neural networks. Keywords: Support vector machine; fault diagnosis; binary decision tree; mine hoist [b][align=center]Mine Hoist Fault Diagnosis Based on SVM-Binary Decision Tree Method ZHANG Lei, XIA Shi-Xiong, NIU Qiang[/align][/b] Abstract: The problems and defects of the existing methods of SVM multi-class classification were analyzed. A multi-class classification based on binary decision tree was proposed, which is simple and has few duplicating training samples. The application to fault diagnosis for mine hoist fault diagnosis shows that it has higher diagnosis accuracy than traditional SVM multi-class method and BP neural network method. Keywords: support vector machine; fault diagnosis; binary decision tree; mine hoist 1 Introduction In the field of fault diagnosis, one of the main challenges is the discovery of fault feature knowledge, among which fault mode classification is the core of the diagnostic process. Currently, most conventional fault diagnosis methods rely on statistical characteristics under large sample conditions. When the training samples are limited, it is difficult to guarantee good classification generalization. The emergence of Statistical Learning Theory (SLT) and Support Vector Machines (SVM) has opened up new avenues for solving this problem. Statistical Learning Theory, based on the principle of minimizing structural risk, is a new theoretical system specifically designed for machine learning problems with small sample sizes. Statistical inference within this system does not aim for the optimal solution when the number of samples approaches infinity, but rather seeks the optimal solution with a finite number of samples, representing a trade-off between empirical risk and confidence range. Support Vector Machines are a new machine learning algorithm developed based on Statistical Learning Theory; they are a concrete application of SLT theory. This paper proposes a multi-class classification method based on binary decision trees using SVM and applies it to fault diagnosis of a hoist system. The method is compared with BP neural networks and traditional multi-class SVM algorithms. Experiments demonstrate that this method not only has excellent classification ability for fault diagnosis of mine hoists with small sample sizes but also exhibits good adaptability and computational efficiency. 2 Related Work With the increasing automation and reliability requirements of hoist operations, research on fault diagnosis technology is of significant positive importance. Existing research on fault diagnosis for hoist systems mainly focuses on methods based on artificial neural networks and traditional multi-class support vector machines. Literature studies the use of artificial neural networks for safety monitoring and fault diagnosis of hoist braking systems. Artificial neural networks essentially use gradient descent to adjust weights to minimize the objective function, leading to an overemphasis on overcoming learning errors and resulting in weak generalization performance. Furthermore, neural networks have some insurmountable drawbacks, such as the difficulty in determining the number of hidden layer units and the significant influence of initial values ​​on the final weights. Neural network classifiers often face the challenge of obtaining a decision function with high generalization ability from a limited number of fault samples. For details, please click: Fault Diagnosis of Mine Hoists Based on SVM Binary Decision Tree Method