Under the background of big data development, students have diverse and multi-level demands for music aesthetic education teaching. This paper studies the implementation strategy of aesthetic education in college music teaching under the threshold of big data, mainly through data analysis, predictive modeling and simulation, multi-faceted analysis, and the design of personalized music teaching intervention. By training DNN model experimental conclusion and analysis, the DNN model in the original set R-squared reached 0.8037, the largest performance among all models, compared to multiple linear regression, in multi-level multi-neuron activation, through the sigmoid activation of each level and linear calculation between levels, in backpropagation to update the weights and bias, trained to be able to more The DNN is finally selected as the best model for music learners’ performance prediction. The DNN is finally selected as the best model for music learners’ performance prediction. The advantages of big data mining music aesthetic education resources are exploited to create an open aesthetic education teaching space for students.