During the cutting process of end milling cutters, the cutting surface will wear out, and the degree of wear is usually divided into three stages: initial wear stage, normal wear stage, and severe wear stage. Obviously, the sampling data in the normal wear stage is much larger than that in the initial wear stage and severe wear stage, resulting in an imbalance in the data set under the tool wear label, which will reduce the accuracy of the deep learning network model in predicting the tool wear state. To address this issue, this article proposes an enhanced method for tool wear condition monitoring dataset based on improved GAAE, which leverages the reconstruction accuracy of GAAE and the sample control ability of cGAN to fully exploit the advantages of both models. The sensor collect the vibration signal during the milling process, convert the vibration signal into spectral data and input them into GAAE. GAAE learns the data distribution characteristics through the autoencoder to generate initial sample data of the tool wear state. The generated samples are input together with the condition vector into the discriminator of cGAN. The discriminator of cGAN further distinguishes between generated samples and real samples, and introduces the condition vector to identify the specific characteristics or attributes of the samples. Afterwards, the enhanced dataset is input into a deep learning network model for classification, testing the usability of the generated data. The experimental results show that training a deep learning network model with enhanced tool wear state data sets can effectively improve the accuracy of the model for tool wear state monitoring, with a prediction accuracy of 96.7%.