Implement of a dengue fever epidemic forecasting web application
Purpose
To prevent the spread of dengue fever, it is necessary to share the awareness of the crisis and to take the vaccine. Information on the current number of infected people and epidemic forecasting systems for each of the four dengue serotypes are necessary for sharing the awareness of the crisis and for the implementation of the vaccine. Information on the current number of cases is regularly published in most countries where dengue is endemic, but few countries have implemented an epidemic forecasting system. Therefore, Dry lab has developed an epidemic forecasting system for each of the four dengue serotypes. In addition, to obtain the latest data essential for the sustainable operation of the epidemic forecasting system, we developed a web application to collect information on Inspection result from the hardware developed by the Wet lab and the Hardware lab.
Development of a dengue fever epidemic forecasting system
Under the circumstances that the number of dengue fever cases is increasing and the epidemic area is expanding, the importance of a prediction system for the number of dengue fever cases is increasing. For example, since the current vaccine is mainly a tetravalent vaccine, the prevalence prediction of "serotypes" is useful for vaccine intake considering the characteristics of dengue fever. Therefore, we developed and evaluated a model to predict the proportion of people infected with four dengue serotypes based on climate data. The climate data used for training were selected to be easily available (average rainfall, average temperature, maximum temperature, and minimum temperature) for ease of use in practice. In addition, we created a web-based system for viewing the forecast results.
The data used were monthly data published by the Singapore health authorities. To understand the epidemic cycle of dengue fever, we compared the impact of climate features in Lasso and Random Forest. The importance of climate features are plotted in Figure 1 and Figure 2, respectively.
The different climatic features that are strongly affected by each serotype are consistent with the facts shown in previous studies.
Using LSTM, which is specialized for learning time series data, we examined the length of the period that influences the serotype infection trend. The accuracy was highest at 6 months.
The final prediction results for the four models (Lasso, Random Forest, Simple Neural Network, and LSTM) are shown in Figure 3, Figure 4, Figure 5, and Figure 6, respectively.
Since these models are able to capture epidemic trends relatively well, we can say that we have achieved our goal of predicting infection trends for each serotype using easily available climate features.
Development web application
Next, we describe a web application developed by Wet lab and Hardware lab that collects inspection data from inspection kits, which is essential for the sustainable operation of the forecasting system, and a web application that displays the results of the forecasting system, which is necessary for sharing crisis awareness.
The reason why we implemented a web application rather than a smartphone application is that it can be used without being bound by the limitations of the device, compared to a smartphone application.
The inspection result of the hardware can be easily posted as shown in the video below.
Inspection results registered via this web application are stored in a database and used for inference of the next and subsequent predictions. The accuracy of the forecasting system improves with the number of data available for training. This means that more people will be able to use more accurate predictions by using the developed hardware and this web application. And they may find new discoveries in the data accumulated through this web application.
The forecasting results of the forecasting system can be easily checked as shown in the video below.
To allow more people to see the prediction results, we have made some innovations to improve the ease of viewing, such as displaying the results in different colors for each of the four serotypes as shown in Figure 7, allowing the user to select some of the four serotypes to view the prediction results as shown in Figure 8, and displaying detailed values by moving the cursor over them as shown in Figure 9 and Figure 10.
