We have developed a detection platform that combines Surface-Enhanced Raman Scattering (SERS) with Machine Learning Algorithms (MLA) for the rapid and sensitive detection of furfural in transformer oil. This is crucial for assessing the degree of aging in transformer oil-paper insulation. Firstly, we synthesized silver nanoparticles (AgNPs) with size-dependent properties using a hydrothermal method. These were then spin-coated onto a gold-plated polycrystalline silicon substrate (Si@Au) to form the Si@Au-Ag SERS substrate. With this substrate, we successfully detected furfural in transformer oil solutions of different concentrations, obtaining the corresponding Raman spectroscopic dataset. Subsequently, we employed two different MLAs, namely PCA+ANN and ANN, to construct quantitative calibration curves, enabling the conversion of detected Raman spectroscopic data into corresponding furfural concentrations. The regression model we established achieved a correlation coefficient (R) of 0.958, indicating high accuracy of the model.