Large power transformers typically use oil-immersed transformers, and the insulation oil can play a role in heat dissipation and insulation. By analyzing the gas composition in the oil, transformer failures can be predicted. Support vector machines are a data mining algorithm suitable for small sample data, but they are aimed at binary classification problems. However, there are multiple types of transformer failures. Therefore, a multi-label radial basis support vector machine is proposed for transformer failure prediction. The model analyzes gases with strong correlations among various gas content to construct a transformer failure prediction model. Machine learning algorithms can quickly mine potential rules in data, providing references for reasonable maintenance plans for transformers.