With regard to the measured Raman spectral data of transformer insulating oil, the relevant research is carried out based on Raman spectral data processing, feature extraction and aging diagnosis with Raman spectral technology as a means to accurately identify the aging state of transformer insulation. First of all, combined with the law of change of Raman spectral noise, based on the wavelet transform theory, the global threshold wavelet transform filtering method is proposed, which effectively removes the noise signals in Raman spectra. Further, an improved baseline correction method is proposed based on the adaptive iterative reweighting penalized least squares (airPLS) method, which accurately removes the fluorescence background of Raman spectra. Secondly, the aging feature information in Raman spectra is extracted using successive projections algorithm (SPA), and its relationship with the aging degree of transformer insulating oil is analyzed. Finally, the light gradient boosting machine (LightGBM) classification model is used to realize accurate discrimination of transformer insulation aging state, and the extreme gradient boosting (XGBoost) model is used as a control group to compare the diagnostic accuracy of the two. The experimental results show that the LightGBM model possesses obvious advantages in diagnostic accuracy, and also further verifies the validity of the extracted aging feature information.