Online monitoring of gases dissolved in transformer oil is widely applied. Improving the performance of dissolved gas analysis (DGA)-based fault diagnosis methods by exploring new features of time-series data has become an appealing topic. In this study, a new type of correlation features between characteristic gases was extracted from timeseries data based on the maximal information coefficient (MIC), and a fuzzy inference system was established. After the introduction of the principle of the MIC and a method for calculating the MIC-based correlation features, the dominant symptom features that can be used to classify fault types were extracted through the receiver operating characteristic curve. Then, fuzzy rules were learnt, and a fuzzy inference system was designed. In addition, to improve the feasibility of the method, the Newton interpolation method was used for adaptation to the existing sampling cycle. The diagnostic results of the test data show that the proposed method has excellent performance and outperforms some prevailing traditional rule-based methods as well as some artificial intelligent methods. The results also show that by exploring new correlation features from time-series data based on the MIC, the performance of DGAbased methods can be improved.This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.