关键词: 皮肤病诊断, 朴素贝叶斯（NB）, 词频-逆文档频率（TF-IDF）算法, 生成对抗网络（GAN）, 类不平衡数据

Abstract: Different incidences of skin diseases lead to the imbalance of skin diseases data, so that there is a great challenge to train an effective and accurate diagnosis model by machine learning. In this paper, a binary classification diagnosis method is proposed, which is a fusion of GAN (generative adversarial networks) and NB (naive Bayes)    algorithm. In this method, a naive Bayes binary classifier is trained for each disease diagnosis, and a GAN model is innovatively trained to generate more samples for training the naive Bayes binary classifier, bringing positive samples and negative samples to balance. As for multiple classification of skin diseases diagnosis, a multiple classification   diagnosis method is proposed, which is a fusion of GAN and NB algorithm. In this method, a binary classifier is trained by NB algorithm and GAN, with TF-IDF algorithm added in the classifier, and the multiple classifier consists of many binary classifiers. The experiments compare the proposed diagnosis methods with 6 methods. The precision and recall of the two skin diseases diagnosis methods proposed in this paper are improved.

Key words: skin diseases diagnosis, naive Bayes (NB), term frequency-inverse document frequency (TF-IDF) algorithm, generative adversarial networks (GAN), imbalanced data