• Geometric similarity characteristics of matrix cooling structure are investigated. • The prediction of cooling performance is performed by machine learning. • The cooling performance increases exponentially with the increasing scaling factor. Gas turbine blades require high-performance cooling structures to meet the increasing turbine inlet temperature requirements. The matrix cooling structure has high heat transfer performance, but at the same time brings a large resistance loss. In basic research, the cooling structure is usually modeled according to the similarity criterion, but there is doubt about the generalization of the results obtained from the scaled model to the real-size cooling structure in engineering applications. Therefore, numerical simulation was applied to study the similarity characteristics of matrix cooling performance and the effects of geometric scaling factor in this paper. Through data fitting, it was found that the heat transfer performance and resistance performance increased exponentially with the increase of the scaling factor. As geometric scaling factor shrank 10 times, the heat transfer and resistance loss decreased by 21.9% and 34.7% respectively. The heat transfer and resistance loss of each sub-duct were also markedly affected by the geometric scaling. In small scaling factor cases, the heat transfer was weak in the downstream areas of the cooling channel, and there was an obvious gap in resistance loss for each sub-duct. Due to the same geometric characteristics, the flow characteristics of the channels were less affected by geometric scaling factor, but the relative velocity increased with the decrease of the scaling factor. The exponential regression equations of the heat transfer and resistance with respect to the scaling factor were established by machine learning, providing an accurate and appropriate prediction for engineering application.