In response to the limitations of data collection methods, poor data quality, slow collection speed, and inaccurate expression of children’s learning state data in the current generation of children’s learning state data, this chapter applied generative adversarial network (GAN) to the children’s learning state data to solve the existing problems. First, real data on children’s learning were collected, mainly through the installation of monitoring devices in classrooms, and then a generative adversarial network model was constructed. The generator in the model mainly generates realistic children’s learning state data based on the collected real sample data, while the discriminator’s task is to distinguish whether the input data samples are real or generated by the generator. The experimental results showed that the error rate between the generated data of the learning state data generation model studied in this chapter and the real data was below 1.19%. Ten experiments were conducted, and the average error rate of all these experiments was 0.87%. Applying GAN to the generation of children’s learning state data can provide educators and parents with more accurate and comprehensive information on children’s learning state, which helps to better understand and support the children’s learning process.