Objective To address the challenge of coupling the concentrations of H2 CH4 C2 H6 C2 H4 and C2 H2 gases in transformer oil and to improve the accuracy of power transformer fault diagnosis this paper proposed a method using the combination of Stacked Sparse Autoencoder SSAE and XGBoost model to improve the accuracy of power transformer fault diagnosis. Methods Firstly a stacked sparse autoencoder SSAE was used to process DGA data. Secondly the number of stacks in the autoencoder was determined to establish the number of implicit layers. Thirdly SSAE was applied to transform the original data and extract deep-level feature information. Fourthly to address the issue of significant magnitude differences between data the extracted feature data were normalized. Finally the processed data was input into the XGBoost model for classification and verification. Results The fault diagnosis accuracy of the proposed method based on stacked sparse autoencoder and XGBoost for power transformer was 91. 11% which was higher than that of other commonly used machine learning models. Conclusion Experimental results verify the effectiveness of the proposed method demonstrating that this power transformer fault diagnosis method based on stacked sparse autoencoder and XGBoost can effectively improve the accuracy of fault diagnosis for power transformers.