Research And Implement Of Algorithm On Predicting Infectious Diseases Based On Dynamic Network Marker

Even in today’s society with a highly developed medical system,infectious diseases that can spread rapidly,such as influenza and COVID-19,still endanger a huge scale of life.From the perspective of systems biology,the spread and evolution of these infectious diseases often has a critical point,which indicates the mutation and outbreak of infectious diseases.Obviously,detecting the critical point of infectious diseases and further predicting infectious diseases is a key step in the early diagnosis,prevention and control of infectious diseases.In view of the difficulties caused by the complexity of time and space in the evolution and spread of infectious diseases,based on the Dynamic Network Marker(DNM)proposed by Chen Luonan,the article makes the following optimizations and improvements in detecting critical point of infectious disease and predicting infectious disease:1)The shortest-path-based dynamic network marker(SP-DNM)is proposed.This algorithm models the area affected by the epidemics into a dynamic network,and transforms the evolution process of the epidemics into a dynamic change process of the regional network.By calculating the correlation coefficient and other statistical indicators,edges in the network can be measured precisely.After introducing the shortest path in graph theory,the shortest paths of the city network between key nodes provides an accurate way to quantitatively measures state of network.The algorithm digs out the temporal and spatial characteristics of the infectious disease state in the modeled dynamic network,detects critical mutations in the overall network,and issues early warnings for the outbreak of infectious diseases.2)Machine learning epidemics time series prediction model based on SP-DNM is proposed.From the aspect of feature improvement,the SP-DNM algorithm is used to extract dynamic spatio-temporal features representing the change of infectious disease status from basic time series information and regional geographic location information,so that the model can more easily capture the development law of infectious diseases;At the same time,XGBoost and Light GBM are used as the basic model training respectively,which means the efficient and convenient characteristics of the integrated learning algorithm are used to improve the accuracy of prediction.Finally,the two basic models are merged to make the development of infectious diseases more accurate.To demonstrate the effectiveness of our method from the perspective of different regions and different diseases,we applied it to the historical records in Tokyo and Hokkaido for consecutive 9 years,from 2010 to 2019 in Tokyo and from 2009 to 2018 in Hokkaido,as well as the COVID-19 data in Kanto,Japan,2020.Comparing with traditional DNM algorithm,SPDNM accurately predicted each infectious disease outbreak and get higher stability,which is helpful for research on the spread of infectious diseases.By using the SP-DNM time series prediction model,the predicted curve obtained is more similar to the actual result comparing with traditional model,which reduces the error and has made a big improvement.Finally,based on the proposed algorithm,an infectious disease forecast system is designed and implemented for practical application,which can detect the early-warning of influenza or COVID-19 in Tokyo and other region,which is of great significance to infectious disease prevention in the future.