Breast cancer is the first malignant tumor in women worldwide. Studying on breast cancer diagnosis and prediction methods based on neural network models is to combine clinical and machine learning to help medical workers more quickly and accurately determine the disease or not, and solve the problems of over-fitting, missed diagnosis rate and high misdiagnosis rate in existing models, and improve the accuracy of prediction models. The University of California Irvine(UCI) data set contains 669 samples, including 357 benign samples and 212 malignant tumor samples, a total of 10 features to train the prediction model. The 10 neural network models are combined through Adaboost method, that is, multiple weak classifiers are combined by Adaboost algorithm to form a strong classifier. The final output is an integrated prediction model with higher accuracy, stronger self-learning ability, adaptive ability and excellent generalization performance. The conclusion shows that the prediction accuracy of the model is 98.550 7%, and the Accuracy(AUC) is 0.996 6, which indicates that the established model is very stable, and has good discrimination and good verification effects. It provides further technical support and guarantee for clinical application.