| Dengue fever is an acute infectious disease caused by dengue virus transmitted by Aedes aegypti mosquitoes,with the characteristics of rapid transmission and high incidence.In recent years,the dengue fever epidemic in Guangzhou has become increasingly serious and has become a high incidence area in China,and taking appropriate prevention and control measures based on the transmission pattern of dengue fever is an important means to prevent its outbreak.At present,studies on the spatial and temporal distribution of dengue epidemics are mostly conducted on a large scale,which can no longer meet the realistic needs of prevention and control.The use of grid units for dengue transmission risk prediction breaks the restrictions of administrative boundaries,facilitates cross-administrative collaboration to prevent dengue transmission,and to a certain extent can provide important methodological references for local disease control departments to develop more reasonable solutions.In this paper,we collected dengue fever case data from 2017-2019 in Guangzhou city,combined with natural and socio-economic data such as precipitation,surface temperature,population density,road density,normalized vegetation index,hospital accessibility,bus stop density and Shannon uniformity index of land use,and processed them using an oversampling method to solve the problem of sample imbalance.A random forest model was used to spatially predict the risk of dengue fever epidemic transmission at a scale of 1km×1km.The main research contents and conclusions of this paper are as follows:(1)The spatial distribution,temporal distribution and population distribution were used to analyze the spatial and temporal distribution characteristics of dengue cases.The results showed that the number of dengue fever cases in Guangzhou City during 2017-2019 showed an increasing trend year by year.Dengue fever was prevalent in June-November,with peak incidence in August-October,peak incidence in the age group between 18-65 years old,and the occupation of patients was dominated by household and non-working,commercial services,workers,retired workers,and student groups.The spatial distribution showed that dengue fever has a high aggregation,forming different aggregation centers.The main aggregation centers were all the streets in Yuexiu District,some streets in Liwan District and Haizhu District,and the secondary aggregation centers were distributed in areas such as Huangcun Street and Zhuji Street in Tianhe District and Yuzhu Street in Huangpu District.The spatial distribution of dengue fever is expanding year by year,and the degree of aggregation is decreasing year by year.(2)The samples are preprocessed and the sample balancing process is performed using the oversampling method.After aggregating the case data statistics to the 1km×1km grid,the number of case-free grids will be much more than the case-free grids,and there is a sample distribution imbalance problem,which leads to poor results of the machine learning model.In this paper,three resampling methods of under-sampling,oversampling and combined sampling are used for processing,and the prediction results based on the three methods are compared with the modeling results of the original data,and the data after the oversampling process is selected for random forest model prediction.The final test AUC value of 0.9995,accuracy of0.9783,precision of 0.9989,recall of 0.9589,and F1 value of 0.9790 were obtained,which were improved by 4.23%,45.41%,22.41%,35.35%,and 5.53%,respectively,compared with the modeling results of the original data.(3)The relationship between univariate factors and the risk of dengue transmission and the degree of importance for dengue prediction were analyzed.In this paper,a total of eight influencing factors were selected for the prediction model after considering the univariate prediction model to test the AUC values and the correlation between the variables.The increase in mean square error(%Inc MSE)was used as an indicator of the importance of the variables,and it was found that population density was much more important than other variables,with the %Inc MSE of 63.76.Hospital accessibility was the second important characteristic variable,and mean surface temperature was the least important among the selected variables,with the %Inc MSE of 35.42.The bias dependence plot in the random forest model was used to show the relationship between the influencing factors and dengue fever.The influence of NDVI on the spread of dengue epidemic was in the shape of a “V”,which was decreasing and then increasing.Areas with high population density,road density,bus stop density and land use type with high SHEI index increased the risk of dengue transmission in Guangzhou.Average rainfall and surface temperature in a certain range increase the risk of dengue transmission.Hospital accessibility within 20-60 min would reduce the risk of transmission of the epidemic.(4)The prediction results of dengue transmission risk showed that the risk of dengue transmission in Guangzhou was mainly distributed in the central city of Guangzhou,centered on Yuexiu,Liwan and Haizhu districts,extending northward to the middle of Baiyun district,southward to the junction of Panyu and Nansha districts,and eastward to the eastern part of Huangpu district.The risk areas in the prediction results were highly consistent with the distribution of cases,indicating that the method proposed in this study can accurately portray the geographical distribution of dengue transmission risk.The distribution of epidemic risk in Guangzhou was consistent with the distribution of high-density population areas,with the area of high-risk areas accounting for 6.48% of the total area and the population at risk accounting for 39.13% of the total population.Four districts,Yuexiu District,Liwan District,Haizhu District and Tianhe District,have more than 80% of the population in the high-risk area and can be considered for comprehensive prevention and control.Meanwhile,key prevention and control is carried out in some townships in Baiyun District,Huangpu District and Panyu District. |
PDF Full Text Request