Detection And Prediction Algorithm For Flu Based On Social Network

Researches on big Data are changing the world. Social network is a population which creating large data with hybrid structure. Research on big data is affecting people’s life. The detection and control of the epidemic is one of the important aspect of public health. At present, clinical data is the main data to support the research of epidemic disease detection and control. For the data latency in clinical data, it’s hardly to advanced control and predict the epidemic. However, the social network data brings hope to early detection and prediction. In recent years, researches based on social network spring up. Event predictions with social data have good effects. Effective processing methods on social data can provide early detection and prediction of epidemics which avoid overmuch material resource cost caused by the spread of epidemics.In this paper, we focused on the effective methods of detection and prediction of flu with social data. Firstly, we use ELI keywords to filter the source data finding items including ILI symptom keywords. For items including keywords are not necessarily describing the user’s symptom, hence classification algorithms including KNN, Naive Bayes and SVM are conducted on the filtered data by which finding a better classification result as our core research data to support the method for flu detection and prediction. To evaluate the social data associating with CDC data, we use incidence to measure the two source data, and PM2.5 as supporting data, and finally find a positive correlation between social and CDC incidence. To predict the transmission of flu, we adopt the Dynamic Bayesian network and the Hidden Markov model, in which social network, geographic position and the PM2.5 in the past week are considered as the impact factors. With repeat cross-over experiments, the model for prediction the flu are verified.