Application of Data Mining Technology on Surveillance Report Data of HIV/AIDS High-Risk Group in Urumqi from 2009 to 2015

• Published in: Complexity (New York, N.Y.) N.Y.), 2018-01, Vol.2018 (2018), p.1-17
• Main Authors: Zheng, Yujian, Wang, Kai, Li, Huling, Yang, Fang, Zhang, Xueliang, Xu, Jiabo, Zhang, Man, Tang, Dandan, Feng, Li
• Format: Article
• Language: English
• Subjects: Accuracy; Acquired immune deficiency syndrome; AIDS; Algorithms; Artificial intelligence; Breast cancer; Classification; Data mining; Decision trees; Demographics; Diagnostic systems; Disease; Epidemics; Fatalities; Forecasts and trends; Growth rate; Health aspects; Health surveillance; HIV; HIV infection; Human immunodeficiency virus; Infections; Mathematical models; Medical personnel; Medical research; Neural networks; Performance prediction; Public health; Risk; Support vector machines
• ISSN: 1076-2787; 1099-0526
• DOI: 10.1155/2018/9193248

Objective. Urumqi is one of the key areas of HIV/AIDS infection in Xinjiang and in China. The AIDS epidemic is spreading from high-risk groups to the general population, and the situation is still very serious. The goal of this study was to use four data mining algorithms to establish the identification model of HIV infection and compare their predictive performance. Method. The data from the sentinel monitoring data of the three groups of high-risk groups (injecting drug users (IDU), men who have sex with men (MSM), and female sex workers (FSW)) in Urumqi from 2009 to 2015 included demographic characteristics, sex behavior, and serological detection results. Then we used age, marital status, education level, and other variables as input variables and whether to infect HIV as output variables to establish four prediction models for the three datasets. We also used confusion matrix, accuracy, sensitivity, specificity, precision, recall, and the area under the receiver operating characteristic (ROC) curve (AUC) to evaluate classification performance and analyzed the importance of predictive variables. Results. The final experimental results show that random forests algorithm obtains the best results, the diagnostic accuracy for random forests on MSM dataset is 94.4821%, 97.5136% on FSW dataset, and 94.6375% on IDU dataset. The k-nearest neighbors algorithm came out second, with 91.5258% diagnostic accuracy on MSM dataset, 96.3083% diagnostic accuracy on FSW dataset, and 90.8287% diagnostic accuracy on IDU dataset, followed by support vector machine (94.0182%, 98.0369%, and 91.3571%). The decision tree algorithm was the poorest among the four algorithms, with 79.1761% diagnostic accuracy on MSM dataset, 87.0283% diagnostic accuracy on FSW dataset, and 74.3879% accuracy on IDU. Conclusions. Data mining technology, as a new method of assisting disease screening and diagnosis, can help medical personnel to screen and diagnose AIDS rapidly from a large number of information.