Human Mobility-based Individual-level Epidemic Simulation Platform
Authors: 
Zipei Fan, 
Chuang Yang, Zhiwen Zhang, Xuan Song, Yinghao Liu, 
Renhe Jiang, Quanjun Chen, Ryosuke ShibasakiAuthors Info & Claims
Zipei Fan, Chuang Yang, Zhiwen Zhang, Xuan Song, Yinghao Liu, Renhe Jiang, Quanjun Chen, Ryosuke ShibasakiAuthors Info & ClaimsArticle No.: 19, Pages 1 - 16
Abstract
COVID-19 has spread worldwide, and over 140 million people have been confirmed infected, over 3 million people have died, and the numbers are still increasing dramatically. The consensus has been reached by scientists that COVID-19 can be transmitted in an airborne way, and human-to-human transmission is the primary cause of the fast spread of COVID-19. Thus, mobility should be restricted to control the epidemic, and many governments worldwide have succeeded in curbing the spread by means of control policies like city lockdowns. Against this background, we propose a novel fine-grained transmission model based on real-world human mobility data and develop a platform that helps the researcher or governors to explore the possibility of future development of the epidemic spreading and simulate the outcomes of human mobility and the epidemic state under different epidemic control policies. The proposed platform can also support users to determine potential contacts, discover regions with high infectious risks, and assess the individual infectious risk. The multi-functional platform aims at helping the users to evaluate the effectiveness of a regional lockdown policy and facilitate the process of screening and more accurately targeting the potential virus carriers.
1 Introduction
Human mobility is a key factor in analyzing the spreading process of a contagious disease [16, 25], especially for COVID-19 [11, 28, 29], which is highly contagious even before symptoms show up. Thus, to quickly identify the potential patients and take measures to prevent the spreading of the disease in the earliest stage has been proven to be the most effective way to slow down the spreading speed and reduce the outbreak risk of the disease.
Sensing human mobility at a large scale and at the individual level is critical to more fine-grained epidemic control. Thanks to the popularization of devices with an accurate localization function and a more powerful and ubiquitous wireless network, sensing human mobility over a very long time span (months or years) and at a large geographical scale is possible, and many trajectory databases that record the movements of the subjects have been made available to researchers worldwide. With the help of such datasets, we can gain a more clear insight into how people move in a city with a fine spatial and temporal granularity, which supports a finer analysis and simulation of the epidemic spreading.
Moreover, the conventional epidemic model (e.g., Susceptible, Infectious, and Recovered (SIR) model [3] or Susceptible, Exposed, Infectious, and Recovered (SEIR) model [20]) predicts the future trends of epidemic spread at a coarse level, where only the infected/susceptible/ recovered/exposed population at a citywide/nationwide level is simulated. The limitations can be summarized from the following three aspects:
•
The contact ratio is assumed to be unchanged during the simulation. However, in practice, the contact ratio is quite different from day to day regarding the dynamics of human mobility.
•
The contact ratio is assumed to be uniformed at different locations in the city, while the city/nation is taken as a whole, while the downtown area, where people are more crowded and rural areas, where people are much less crowded, the contact ratios are quite different.
•
The simulation is conducted for aggregated population rather than each individual, hence we can hardly estimate the infected risks of each individual or discover high-risk regions and track those users that visit the high-risk regions.
Recent advances in epidemic simulations and predictions have taken human mobility into consideration. Reference [23] proposed a proxy for individual mobility and used it to describe the flow of people and predict the spread of an influenza-like contagious disease. The proxy utilized in the paper is close to an origin-destination (OD) matrix that is widely in the transportation studies. [26] is a recent study on the human mobility effect in the spreading of COVID-19, but only the OD matrix is used to estimate the spatial transition of human mobility instead of the trajectories of each user. This makes it difficult to find potential contacts and predict the infection risk of each individual.
Moreover, it has been proven that the most effective way to slow down the spread of COVID-19 is to implement some public policies such as city lockdown, telework, or setting temperature check points. However, all these public policies are “expensive” and may cause severe political risks and financial damage to society. Thus, it is important for governments to know how effective a public policy will be before taking any action and guarantee that public policies are worth the potential risks. Against this background, we aim at simulating human mobility in response to public policies and analysing how effective public policies can slow down the epidemic spreading process.
In this article, we develop a novel epidemic simulation platform that provides three principal functions: (1) simulate the epidemic spreading on the real-world human trajectory dataset, where people contacting behaviors are estimated by GPS trajectories; (2) simulate human mobility in response to the public policies and simulate and compare the effectiveness of each policy by conducting the epidemic spreading simulation on the simulated human mobility; and (3) support different data exploration and data mining on the infection spreading risks, such as discovering the primary/secondary potential contacts and regions with high infection risk. The user interface of this epidemic simulation platform is shown in Figure 1.
Fig. 1.
In the rest of this article, we will review existing studies related to our work, and introduce our human mobility-based individual-level epidemic simulation platform focusing on three functions: (1) probabilistic individual-level contagious disease transmission model, (2) trajectory replacement-based public policy simulation, and (3) regional and individual infection risk exploration.
2 Related Work
Epidemiological models have been proposed in the early of 20th century [15], and many extensions have been made from different aspects. The conventional epidemiological SIR model [15] comprises of three labels of populations: S, I, and R (Susceptible, Infectious, and Recovered). To model the epidemic disease without immunity, Reference [19] proposed a SIS model that removes the “recovered” label and the infectious population can only turn into susceptible. The SEIR model [20] introduces a new label “E” (exposed) that considers the incubation period during which individuals have been infected but are not yet infectious themselves. Vaccination is taken into consideration in Reference [10] by extending the recovered population by concerning the number of vaccinated subjects. In particular, SEIR has been considered the fundamental model that is suitable for COVID-19 [5, 33]. Thus, in this article, we use the SEIR model and adapt it to simulate the epidemic at the individual level; note that our simulation model can be easily adapted to different epidemic models such as SIS or Vaccinated SEIR by changing Equation (1).
There are many emerging studies [7, 9, 16, 23, 26] on analyzing the relationship between human mobility and the spreading of infectious disease, especially since the outbreak of COVID-19, which is highly contagious and is contagious even during the incubation period. Reference [17] models the spreading of COVID-19 using real-time human mobility data obtained from Baidu Inc. From the simulation, the authors found that local control measures such as isolation instead of long-range travel restriction play a more important role in curbing the spreading process. Reference [27] and Reference [6] leverage Google Mobility Data1 to find out the relationship between the spreading of COVID-19 and human mobility. Reference [5] provides a meta-population SEIR model to predict the future epidemic spreading concerning different racial and social-economic groups. Reference [14] provides a good survey of how human mobility can be utilized to analyse COVID-19. Most of these existing studies predict or simulate human mobility at the aggregated population or origin-destination matrix level, while the individual level is seldom taken into consideration. Several recent studies [18, 21, 22] concern individual trajectories to develop agent-based epidemic simulation system. Agent-based simulation is conducted based on sophisticated assumptions and may not reflect real-world human mobility well if the simulation is not well configured. In this article, we utilize real-world GPS trajectories of users and conduct trajectory replacement strategies to simulate human mobility in response to different lockdown policies, which is a more data-driven approach that requires fewer assumptions and can be more simply configured. Besides, these previous studies focus on analysing the epidemic simulation per se but fewer data exploration functions that support an interactive simulation of different lockdown policies or data exploration on the simulation results to discover regions or individuals with high risk.
Some previous studies on visual analytic on epidemic simulation and control also inspire this work. Reference [13] proposes a visual analytic approach that synthesizes very large spatial interaction data and discovers interesting patterns, which support the decision-making process during the pandemic. References [1, 2] measure different control policies via an interactive user interface and compare the effectiveness of each control policy. Reference [8] visualizes the spread of disease in relation to population density and other demographic conditions and supports the user to discover both short- and long-range epidemic spreading patterns. Compared with these existing studies, our work can provide a more fine-grained epidemic simulation (at the individual level) and discover the risky regions or individuals based on the full trajectories of the subjects.
3 Methods
In this section, we will elaborate our design of individual-level epidemic simulation platform, as shown in Figure 2. First, a trajectory-based compartmental model is proposed to simulate the stochastic disease transmission process, which lays the foundation of our higher-level simulation and infection risk exploration, which will be elaborated in the following subsections, respectively.
Fig. 2.
3.1 Trajectory-based Compartmental Model
In this section, we elaborate the details of how we extend the conventional SEIR model to the trajectory-based compartmental model, which comprises of two phases: (1) grid infection state update and (2) grid state diffusion. Rather than modeling the entire population at the citywide/ nationwide level, we divide the city into grids and run the SEIR model at each grid independently. In our experiments, we utilize Uber’s Hexagonal Hierarchical Spatial Index (H3)2 for geospatial as the grid system, because hexagonal grids are closer to circles thus the distance within the grid shows a better consistency compared with rectangle or triangle shape grids, and hexagonal grids are more convenient for map visualization compared with Geohash3 or S24 [12]. Experimentally, we choose the grid resolution of level 8 (grid size is approximately 0.737 km\(^2\)).
Hence, the conventional SEIR model [33] can be rewritten aswhere \(S_{l,\,t}\), \(E_{l,\,t}\), \(I_{l,\,t}\), and \(R_{l,\,t}\) are the number of susceptible, exposed, infected, and recovered users at location \(l\) at time \(t\). Location \(l\) is represented by the Uber’s H3. Time is sliced into slots with a constant sampling rate of 5 minutes. \(\beta _{1}\) denotes the rate of transmission from susceptible to infected, \(\beta _{2}\) denotes the rate of transmission from susceptible to exposed and \(\gamma\) are the contact rate and recovery rates. Hence the grid state at the next time step can be estimated bywhere \(\Delta t\) denotes the time interval of our simulation. Experimentally, we set the time interval \(\Delta t\) to be 5 minutes (equivalent to \(\frac{1}{288}\) day, concerning that the conventional settings of epidemic parameters \(\beta _1\), \(\beta _2\), \(\gamma\), and \(\sigma\) is on a daily basis). Thus, we can update the infection state of each grid independently at each timestamp. In the next step, we elaborate how we model the grid state diffusion process regarding the mobility of each individual.
\begin{equation} \begin{aligned} \frac{dS_{l,\,t}}{dt} &= -\beta _1 \frac{S_{l,\,t}I_{l,\,t}}{N_{l,\,t}} -\beta _2 \frac{E_{l,\,t}I_{l,\,t}}{N_{l,\,t}} \\ \frac{dE_{l,\,t}}{dt} &= -\sigma E_{l,\,t} + \beta _1 \frac{S_{l,\,t}I_{l,\,t}}{N_{l,\,t}} + \beta _2 \frac{E_{l,\,t}I_{l,\,t}}{N_{l,\,t}} \\ \frac{dI_{l,\,t}}{dt} &= -\gamma I_{l,\,t} + \sigma E_{l,\,t} \\ \frac{dR_{l,\,t}}{dt} &= \gamma I_{l,\,t}, \\ \end{aligned} \end{equation}
(1)
\begin{equation} \Delta S_{l,\,t} = \frac{dS_{l,\,t}}{dt} \Delta t,\quad \Delta E_{l,\,t} = \frac{dE_{l,\,t}}{dt} \Delta t,\quad \Delta I_{l,\,t} = \frac{dI_{l,\,t}}{dt} \Delta t, \quad \Delta R_{l,\,t} = \frac{dR_{l,\,t}}{dt} \Delta t, \end{equation}
(2)
Before we consider the mobility of each individual, we need to update the infection state of each individual based on the grid-level infection state increments. We denote the list of user IDs at location \(l\) at time \(t\) by \(U_{l,\,t}\) and the susceptible, infected, and recovered subsets to be \(U^S_{l,\,t}\), \(U^E_{l,\,t}\), \(U^I_{l,\,t}\), and \(U^R_{l,\,t}\), respectively. From Equation (1), we obtain the simulated incremental numbers \(\Delta S_{l,\,t}\), \(\Delta E_{l,\,t}\), \(\Delta I_{l,\,t}\), and \(\Delta R_{l,\,t}\) at time \(t\) and increase or decrease the lists of users \(U^S_{l,\,t}\), \(U^I_{l,\,t}\), and \(U^R_{l,\,t}\) by sampling new infected users and recovered users with corresponding increments. Note that the increments cannot be ignored, although it is always very small (the spatial and temporal resolution is high, and the incremental number of small areas during a short time period is always very small). However, for simulating the infection process, the numbers of susceptible/exposed/infection/recovered are integers, with a minimal change of 1. Thus, to get the integer values of the increments \(\Delta \hat{S}_{l,\,t}\), \(\Delta \hat{E}_{l,\,t}\), \(\Delta \hat{I}_{l,\,t}\), and \(\Delta \hat{R}_{l,\,t}\), we quantify the incremental numbers by implementing a sampling strategy to use randomness to represent the small values. For example, if our simulated incremental number of infection is 0.001 at location \(l\) at time \(t\), then we sample 1 new infected user from susceptible list \(U^S_{l,\,t}\) with a probability of 0.001.
After obtaining the number of increments, we update the \(U^S_{l,\,t}\), \(U^E_{l,\,t}\), \(U^I_{l,\,t}\), and \(U^R_{l,\,t}\) by randomly (1) sampling the incremental IDs of \(U^R_{l,\,t}\) from \(U^I_{l,\,t}\) with a number of \(\Delta R_{l,\,t}\); (2) sampling the incremental IDs of \(U^E_{l,\,t}\) from \(U^S_{l,\,t}\) with a number of \(\Delta E_{l,\,t}\); (2) sampling the incremental IDs of \(U^I_{l,\,t}\) from \(U^E_{l,\,t}\) with a number of \(\Delta I_{l,\,t}\); and (3) updating \(U^S_{l,\,t}\), \(U^E_{l,\,t}\), \(U^I_{l,\,t}\), and \(U^R_{l,\,t}\) by the above three draws.
To consider the diffusion effect caused by human mobility, we make another update to the user lists \(U^S_{l,\,t}\), \(U^E_{l,\,t}\), \(U^I_{l,\,t}\), and \(U^R_{l,\,t}\) based on the trajectory of each individual. We copy all user lists from time \(t\) to time \(t + 1\) for initialization and then for each user \(u \in U^{c}_{l_t, t}\), where \(c \in \left\lbrace S, E, I, R \right\rbrace\) is the label of infection state and \(l_t\) is the location of user \(u\) at time \(t\), we add the user \(u\) to the next step user list \(U^{c}_{l_{t+1}, t+1}\), and remove user \(u\) from \(U^{c}_{l_{t}, t + 1}\). In this way, we move the user \(u\) from \(l_{t}\) to \(l_{t+1}\) recording his/her infection state \(c\).
After the probabilistic transmission process and mobility-based updates to \(U^S_{l,\,t}\), \(U^E_{l,\,t}\), \(U^I_{l,\,t}\), and \(U^R_{l,\,t}\), we obtain the infection states for the next timestep simulation \(U^S_{l,\,t + 1}\), \(U^E_{l,\,t + 1}\), \(U^I_{l,\,t + 1}\), and \(U^R_{l,\,t + 1}\), which is summarized in Algorithm 1. In the next step, we collect the \(U^S_{l,\,t}\), \(U^S_{l,\,t}\), \(U^I_{l,\,t}\), and \(U^R_{l,\,t}\) for each timestep \(t\) as our simulation results and visualize these users trajectories with their infection state in out platform. As shown in Figure 3, the red markers indicate the infected users from our simulation results, the dark blue markers indicate the susceptible users, and the green markers indicate the recovered users. From Figure 3, we can see the initial infected users (which are generated randomly and do not indicate the real-world infections) that appear on the left on the map and then move to the central area of Shenzhen as shown in the middle of Figure 3. After the infected user come to the central area, where the population density is higher, the spreading of contagious disease is accelerated (as shown on the right of Figure 3).
Fig. 3.
Note that our epidemic simulation is not limited to COVID-19 but a general contagious disease. As shown in Figure 4, we conduct the human mobility-based probabilistic transmission simulation under different parameter (\(\beta _1\) and \(\beta _2\)) settings. The curves show the accumulated number of infected people in each simulation. Algorithm 1 shows that our simulation is conducted with uncertainty, and thus we draw the uncertainty region in Figure 4 based on 50 repetitive simulations.
Fig. 4.
3.2 Lockdown Simulation
To simulate the effectiveness of a regional lockdown policy, in this article we proposed a simple but effective approach named trajectory replacement-based simulation. The basic idea of trajectory replacement-based simulation is to replace the trajectory that does not satisfy the current policy with a history trajectory that satisfies the lockdown policy. As shown in Figure 5, the user first delineates the starting date (we can see from the time slider) and an interested lockdown region on the map, which enforces a restriction on the subjects that cannot move into the delineated region from outside or move from inside to outside. Those trajectories that are passing through or moving within the lockdown region are also restricted. Thus, we find out those trajectories that do not satisfy these restrictions via a spatial join operation, remove those unsatisfied trajectories from the database, and replace them with another trajectory from the same user on different days. If no such trajectories can be found from the users’ historical trajectory database, then we assume that the users stay at home for a whole day.
Fig. 5.
After specifying the starting date and the lockdown region and generating the simulated trajectories via the trajectory replacement strategy, we conduct the probabilistic transmission simulation on both the original trajectories and the simulated trajectories. The effectiveness is compared on the accumulated infected numbers, a comparative view of which is shown on the bottom right corner of Figure 5. From this comparative chart, we can clearly observe how the lockdown policy will work on curbing the spread of the disease. More comparisons of simulations of different lockdown regions (each district in Shenzhen) and different starting days of lockdown policies (since the 1st, 5th, 10th, 15th, 20th, and 25th day) are given in Figure 1. We can clearly see from Figure 6 that an earlier lockdown policy can be more effective in curbing the spread of the epidemic, especially in the central districts in Shenzhen (e.g., Longgang and Futian). This has been proven by a variety of successful and unsuccessful examples in the current COVID-19 pandemic. The countries or cities that take earlier effective mobility restriction policies are usually successfully keep the spreading speed of the epidemic under control and resume normal life earlier.
Fig. 6.
3.3 Infection Risk Exploration
To help the user gain a deeper insight into our simulation results, we implement three functions that support the user to explore the infection risks based on our simulation.
3.3.1 Potential Contacts Tracing.
An effective tracing of contacts is critical in narrowing the potential virus carriers and can significantly facilitate the process of screening or quarantine. Especially, a timely tracing of potential contacts is critical to curbing the epidemic spread. Although a perfect contact tracing is impossible due to the limitation of data accuracy, data sparsity, and the number of users participated, identifying and stratifying potential contacts can still be important in significantly narrowing down the screening range. Aside from the infected users, who have been medically confirmed as infected, those who contacted the infected user within a transmittable distance should also be traced and screened. However, the tracing and screening process is quite costly and time-consuming, especially when taking the secondary and even higher-order contacts into consideration. As shown in Figure 7, as time goes by, the potential contacts ((left) 1 day after the target user is infected and (right) 2 days after the target user is infected) number increase dramatically. Therefore, an efficient tracing that helps the government to accelerate the process of narrowing the potential contacts is critical to minimizing the loss and risk of spreading.
Fig. 7.
To this end, we implement the tracing function that can automatically discover those who may have direct or indirect contact with infected users. The big red marker in Figure 7 represents an infected user (the user is for simulation only and does not correspond to the infection of a real-world user), and the massive small dark blue markers represent the locations of susceptible people. To stratify the contacts into different risk levels, we categorize the contacts into three levels based on the order of the contacts: Primary contacts (level 1) are represented by yellow markers, which indicate that the individuals may have a direct contact with the infected user; secondary/tertiary contacts (levels 2/3) are represented by green/light blue markers, respectively. Individuals with level 1 (primary contacts) risks have direct contact with infected users, which have the highest risk of being infected, thus it is of the highest priority to notify these contacts and conduct a complete screening. Individuals with indirect contacts (secondary or tertiary contacts) are those who have direct contacted with people who are level 1 and level 2, respectively. Although they are also at risk of being infected, the risk is much less than that the primary contacts. Thus, those indirect contacts should also be traced and screened soon after the screening of the primary contacts. In general, the risk of being infected decreases with order of contacts (contact level), while the number of contacts increase exponentially with the contact level. Thus, an accurate categorization of different levels of contacts can be quite helpful in stratifying the strategy of screening and quarantine the potential contacts, especially considering the limit resources of screening.
3.3.2 Tracing Regions with High Infectious Risk.
Identifying clusters of infections and tracing individuals who have visited risky regions play an important role in epidemic control. In the early stage of the COVID-19 outbreak in Wuhan, the Huanan Seafood Wholesale Market has been considered an epicenter of the early outbreak of COVID-19 in Wuhan, and many early infections were found to be closely related to the market. If the individual who visited this market was identified and traced at the earliest time, then the number of secondary, tertiary, and higher-order contacts would have been reduced significantly.
As shown in Figure 8, the regions with high infectious risks can be discovered automatically from our epidemic simulation. Based on the experience and extra knowledge from the experts, we can also manually select or adjust the high-risk regions on the left panel in our platform, and the individuals who visited or passing by these regions will be traced automatically and their movements visualized (shown in red trajectories on the map).
Fig. 8.
3.3.3 Risk Assessment at the Individual Level.
To have be more accurate in identifying and tracing potential infections, we implemented two functions on risk assessment at the individual level. As shown in Figure 9(left), the trajectory of an infected user (the user is for simulation only; it does not correspond to the infection of a real-world user) is visualized. The trajectory before infection is shown in blue, and the trajectory after infection is shown in red. With the help of this visualization, we can easily determine where the individual usually goes and where he/she goes after the infection. This is very helpful in conducting more precise potential contact screening.
Fig. 9.
In Figure 9(right), we show how our platform assesses the infectious risk of each individual. The infectious risk is assessed both from our simulation results and whether the user visited the regions with high infectious risk. The overall risk is indicated by the color of the trajectory, and how many high risk regions the individual has passed through is shown on the left panel in our platform. This function can be helpful in supporting the user to conduct more precise screening, give an explanation of why the user is classified as the high risk person, and identify the “super-spreader” at an early stage.
4 Implementations
4.1 Data
In this work, we use TaxiSZ as a sample of real-world human mobility. TaxiSZ is the taxi dataset used in Reference [34], including 14,728 raw taxi trajectories of Shenzhen on October 22, 2013. The dataset contains 46,927,855 GPS records, which covers the entire Shenzhen area, with a sampling rate of about 13 seconds. Each record contains an anonymized ID, timestamp, latitude, and longitude columns that we can formulate as trajectory. We additionally apply data cleaning on it, due to the inaccuracy of the raw data (e.g., error records). Note that TaxiSZ is only one open sample dataset, any human trajectory dataset with columns of user ID, timestamp, latitude, and longitude with feasible data quality can be directly adapted into our platform. We will discuss on the data selection in details in the next section.
4.2 Platform Implementation
The implementation of our epidemic simulation platform can be divided into two parts: front-end and back-end. On the front-end, the interactive graphical user interface is built under the framework of ReactJS,5 where the geographical data visualization (e.g., regional highlight, trajectory visualization) part is developed with deck.gl [24], and the dynamic updating chart is built with Victory.js.6 On the back-end, the epidemic simulation algorithm is written in Python7 and communicating with interface via a RESTful API built with Flask.8
We deployed our platform on a standalone server with Intel Xeon E5-2690v4 (2.6 GHz 14C 35M 9.60 GT/seconds 135 W), 2\(\times\) TitanX Pascal 12 GB GDDR5X, 128 GB (8 \(\times\) 16 GB DDR4-2400 ECC RDIMM) and 1.2 TB Intel NVMe SSD DC P3600 Series.
5 Discussion
To get more timely simulation results, we expect that our system can finish the simulation within a feasible time. In this spirit, we test the scalability of our system. As shown in Figure 10, our full simulation requires about 30 minutes to conduct one epidemic simulation of one month. We further downsample the dataset with respect to the user ID to see the simulation time growth. A sublinear growth pattern can be observed. This is because the time consumption is mainly twofold: update states by a probabilistic transmission process, where time consumption grows linearly only with the number of nonempty grids (at least one user is within the grid), and update the state by user trajectories, which grows linearly with the number of users. When the user number grows, the nonempty grids grows sublinearly.
Fig. 10.
It is worth noting that our platform is a general platform that can be easily adapted to any type of user trajectory data, such as call detail records (CDR) data [4, 30], GPS data [31], or checkin data [32]. Different types of data will have its pros and cons, for example, traditional CDR data [4] (which only records the nearest base station when the user makes call or sends or receives messages) and checkin data [32] only record the location of a user when the user actively posts on social network with geo-tag) are sparse in nature, this will lead to a significant underestimation of close contacts. When processing with such data, a sophisticated trajectory completion should be conducted to reduce such bias. With the popularization of wireless telecommunication, Internet connection is also recorded in the CDR data [30]. Similarly, a GPS trajectory dataset [31] collected by background applications is also becoming more available to researchers. Such datasets are more suitable to conduct epidemic simulation as well as contact tracing, but such datasets are extremely sensitive to user privacy, and therefore they are very difficult to obtain and the usage of such datasets is often strictly restricted (e.g., cannot visualize the trajectory of a fine resolution and cannot show the home and work locations of the users). Therefore, in this article, we use the open TaxiSZ data, which is both public available as well as most close to the recent CDR data and GPS data in the aspect of sampling rate, to showcase the functions of our system and the rationality of the simulation, and our system users can simply replace the TaxiSZ data with any private dataset (CDR or GPS dataset) as long as each record contains the information of user ID, timestamp, latitude, and longitude.
We also realize several limitations of our current platform: (1) Most of the existing human mobility datasets are samples of real-world human mobility. In our experiments, we also note this problem. To better reflect the real-world infection situation, a scaling factor that indicates each user in the dataset represents how much real-world population should be estimated, and the epidemic transmission parameters should also be adjusted accordingly. (2) Our current contact inference is based on the co-occurrence of two trajectories within the grids. Data sparsity and different functionality of grids have not been taken into consideration. In our future version, we will conduct better trajectory interpolation and adjusting the epidemic transmission parameters with respect to the features of grids, such as POI distribution.
6 Conclusion
In this article, we develop a novel human mobility-based individual-level epidemic simulation platform that simulates the spreading of a contagious disease based on real-world human trajectories data. Rather than simulate the epidemic on the aggregated population, we leverage the human mobility information of each individual and develop a probabilistic transmission model at the individual level. Lockdown policy is simulated via a trajectory replacement strategy, and a comparative view of epidemic simulation with and without the policy is given to give an illustrative view of how effective the policy is. To gain a deeper insight into the simulation results, we develop three functions for infection risk exploration. Potential contacts tracing is implemented to support in narrowing the targets of screening or quarantine and risky individuals, and region mining supports the user to fast estimate the potential risk and identifying potential infections at an early time. These functions are integrated in our epidemic simulation platform and give the user (e.g., disease control and prevention bureaus) a friendly interface to simulate the future possibilities and make better decisions.
Footnotes
4
https://shiba.iis.u-tokyo.ac.jp/member/ueyama/mm/.
References
[1]
Shehzad Afzal, Sohaib Ghani, Hank C. Jenkins-Smith, David S. Ebert, Markus Hadwiger, and Ibrahim Hoteit. 2020. A visual analytics based decision making environment for covid-19 modeling and visualization. In 2020 IEEE Visualization Conference (VIS). IEEE, 86–90.
[2]
Shehzad Afzal, Ross Maciejewski, and David S. Ebert. 2011. Visual analytics decision support environment for epidemic modeling and response evaluation. In Proceedings of the IEEE Conference on Visual Analytics Science and Technology (VAST’11). IEEE, 191–200.
[3]
Roy M. Anderson and Robert M. May. 1992. Infectious Diseases of Humans: Dynamics and Control. Oxford University Press.
[4]
Vincent D. Blondel, Markus Esch, Connie Chan, Fabrice Clérot, Pierre Deville, Etienne Huens, Frédéric Morlot, Zbigniew Smoreda, and Cezary Ziemlicki. 2012. Data for development: The d4d challenge on mobile phone data. arXiv:1210.0137. Retrieved from https://arxiv.org/abs/1210.0137.
[5]
Serina Chang, Emma Pierson, Pang Wei Koh, Jaline Gerardin, Beth Redbird, David Grusky, and Jure Leskovec. 2021. Mobility network models of COVID-19 explain inequities and inform reopening. Nature 589, 7840 (2021), 82–87.
[6]
Wen-Hao Chiang, Xueying Liu, and George Mohler. 2021. Hawkes process modeling of COVID-19 with mobility leading indicators and spatial covariates. International Journal of Forecasting.
[7]
Alana Corsi, Fabiane Florencio de Souza, Regina Negri Pagani, and João Luiz Kovaleski. 2020. Big data analytics as a tool for fighting pandemics: A systematic review of literature. J. Ambient Intell. Human. Comput. (2020), 1–18.
[8]
Cody Dunne, Michael Muller, Nicola Perra, and Mauro Martino. 2015. VoroGraph: Visualization tools for epidemic analysis. In Proceedings of the 33rd Annual ACM Conference Extended Abstracts on Human Factors in Computing Systems (Seoul, Republic of Korea) (CHI EA’15). Association for Computing Machinery, New York, NY, USA, 255–258.
[9]
Solveig Engebretsen, Kenth Engø-Monsen, Mohammad Abdul Aleem, Emily Suzanne Gurley, Arnoldo Frigessi, and Birgitte Freiesleben de Blasio. 2020. Time-aggregated mobile phone mobility data are sufficient for modelling influenza spread: The case of bangladesh. J. Roy. Soc. Interface 17, 167 (2020), 20190809.
[10]
Shujing Gao, Zhidong Teng, Juan J. Nieto, and Angela Torres. 2007. Analysis of an SIR epidemic model with pulse vaccination and distributed time delay. J. Biomed. Biotechnol. 2007 (2007).
[11]
Marino Gatto, Enrico Bertuzzo, Lorenzo Mari, Stefano Miccoli, Luca Carraro, Renato Casagrandi, and Andrea Rinaldo. 2020. Spread and dynamics of the COVID-19 epidemic in Italy: Effects of emergency containment measures. Proc. Natl. Acad. Sci. U.S.A. 117, 19 (2020), 10484–10491.
[12]
Google. [n.d.]. S2. Retrieved from https://s2geometry.io/.
[13]
Diansheng Guo. 2007. Visual analytics of spatial interaction patterns for pandemic decision support. Int. J. Geogr. Inf. Sci. 21, 8 (2007), 859–877.
[14]
Tao Hu, Siqin Wang, Bing She, Mengxi Zhang, Xiao Huang, Yunhe Cui, Jacob Khuri, Yaxin Hu, Xiaokang Fu, Xiaoyue Wang, et al. 2021. Human mobility data in the COVID-19 pandemic: characteristics, applications, and challenges. International Journal of Digital Earth 14, 9 (2021), 1126–1147.
[15]
William Ogilvy Kermack and Anderson G. McKendrick. 1927. A contribution to the mathematical theory of epidemics. Proc. Roy. Soc. Lond. Ser. A 115, 772 (1927), 700–721.
[16]
Mathew V. Kiang, Mauricio Santillana, Jarvis T. Chen, Jukka-Pekka Onnela, Nancy Krieger, Kenth Engø-Monsen, Nattwut Ekapirat, Darin Areechokchai, Preecha Prempree, Richard J. Maude, et al. 2021. Incorporating human mobility data improves forecasts of dengue fever in thailand. Sci. Rep. 11, 1 (2021), 1–12.
[17]
Moritz U. G. Kraemer, Chia-Hung Yang, Bernardo Gutierrez, Chieh-Hsi Wu, Brennan Klein, David M. Pigott, Louis Du Plessis, Nuno R. Faria, Ruoran Li, William P. Hanage, et al. 2020. The effect of human mobility and control measures on the COVID-19 epidemic in China. Science 368, 6490 (2020), 493–497.
[18]
Sebastian A. Müller, Michael Balmer, William Charlton, Ricardo Ewert, Andreas Neumann, Christian Rakow, Tilmann Schlenther, and Kai Nagel. 2021. Predicting the effects of COVID-19 related interventions in urban settings by combining activity-based modelling, agent-based simulation, and mobile phone data. PloS one 16, 10 (2021), e0259037.
[19]
Roni Parshani, Shai Carmi, and Shlomo Havlin. 2010. Epidemic threshold for the susceptible-infectious-susceptible model on random networks. Phys. Rev. Lett. 104, 25 (2010), 258701.
[20]
Kiesha Prem, Yang Liu, Timothy W. Russell, Adam J. Kucharski, Rosalind M. Eggo, Nicholas Davies, Stefan Flasche, Samuel Clifford, Carl A. B. Pearson, James D. Munday, et al. 2020. The effect of control strategies to reduce social mixing on outcomes of the COVID-19 epidemic in wuhan, China: A modelling study. Lancet Publ. Health (2020).
[21]
Sirisha Rambhatla, Sepanta Zeighami, Kameron Shahabi, Cyrus Shahabi, and Yan Liu. 2022. Toward Accurate Spatiotemporal COVID-19 Risk Scores Using High-Resolution Real-World Mobility Data. ACM Transactions on Spatial Algorithms and Systems (TSAS) 8, 2 (2022), 1–30.
[22]
Petrônio C. L. Silva, Paulo V. C. Batista, Hélder S. Lima, Marcos A. Alves, Frederico G. Guimarães, and Rodrigo C. P. Silva. 2020. COVID-ABS: An agent-based model of COVID-19 epidemic to simulate health and economic effects of social distancing interventions. Chaos Solitons Fractals 139 (2020), 110088.
[23]
Michele Tizzoni, Paolo Bajardi, Adeline Decuyper, Guillaume Kon Kam King, Christian M. Schneider, Vincent Blondel, Zbigniew Smoreda, Marta C. González, and Vittoria Colizza. 2014. On the use of human mobility proxies for modeling epidemics. PLoS Comput. Biol. 10, 7 (2014).
[24]
Uber. 2016. deck.gl: WebGL2 Powered Geospatial Visualization Layers. Retrieved from https://deck.gl/.
[25]
Srinivasan Venkatramanan, Adam Sadilek, Arindam Fadikar, Christopher L. Barrett, Matthew Biggerstaff, Jiangzhuo Chen, Xerxes Dotiwalla, Paul Eastham, Bryant Gipson, Dave Higdon, et al. 2021. Forecasting influenza activity using machine-learned mobility map. Nat. Commun. 12, 1 (2021), 1–12.
[26]
Alessandro Vespignani. 2010. Multiscale mobility networks and the large scale spreading of infectious diseases. In APS March Meeting Abstracts, Vol. 2010. A4–002.
[27]
Haiyan Wang and Nao Yamamoto. 2020. Using a partial differential equation with google mobility data to predict COVID-19 in arizona. Math. Biosci. Eng. 17, 5 (2020).
[28]
Yixuan Wang, Yuyi Wang, Yan Chen, and Qingsong Qin. 2020. Unique epidemiological and clinical features of the emerging 2019 novel coronavirus pneumonia (COVID-19) implicate special control measures. J. Med. Virol. 92, 6 (2020), 568–576.
[29]
Joseph T. Wu, Kathy Leung, and Gabriel M. Leung. 2020. Nowcasting and forecasting the potential domestic and international spread of the 2019-nCoV outbreak originating in wuhan, China: A modelling study. Lancet 395, 10225 (2020), 689–697.
[30]
Wenchao Wu, Jiayi Xu, Haipeng Zeng, Yixian Zheng, Huamin Qu, Bing Ni, Mingxuan Yuan, and Lionel M. Ni. 2016. TelCoVis: Visual exploration of co-occurrence in urban human mobility based on telco data. IEEE Trans. Vis. Comput. Graph. 22, 1 (2016), 935–944.
[31]
Takahiro Yabe, Kota Tsubouchi, and Yoshihide Sekimoto. 2017. CityFlowFragility: Measuring the fragility of people flow in cities to disasters using GPS data collected from smartphones. Proc. ACM Interact. Mobile Wear. Ubiq. Technol. 1, 3 (2017), 1–17.
[32]
Dingqi Yang, Daqing Zhang, Zhiyong Yu, and Zhiwen Yu. 2013. Fine-grained preference-aware location search leveraging crowdsourced digital footprints from LBSNs. In Proceedings of the 2013 ACM International Joint Conference on Pervasive and Ubiquitous Computing. 479–488.
[33]
Zifeng Yang, Zhiqi Zeng, Ke Wang, Sook-San Wong, Wenhua Liang, Mark Zanin, Peng Liu, Xudong Cao, Zhongqiang Gao, Zhitong Mai, et al. 2020. Modified SEIR and AI prediction of the epidemics trend of COVID-19 in China under public health interventions. J. Thorac. Dis. 12, 3 (2020), 165.
[34]
Desheng Zhang, Juanjuan Zhao, Fan Zhang, and Tian He. 2015. UrbanCPS: A cyber-physical system based on multi-source big infrastructure data for heterogeneous model integration. In Proceedings of the ACM/IEEE 6th International Conference on Cyber-Physical Systems (ICCPS’15). Association for Computing Machinery, New York, NY, 238–247.
Index Terms
- Human Mobility-based Individual-level Epidemic Simulation Platform
Recommendations
Human mobility based individual-level epidemic simulation platform
Coronavirus has spread worldwide and about 3.5 million people have been confirmed infected and over 300 thousands people died. Scientists have reached the consensus that human mobility is the principal factor in spreading the virus and mobility should be ...
A Cluster-Based Epidemic Model for Retweeting Trend Prediction on Micro-blog
DEXA 2015: Proceedings, Part I, of the 26th International Conference on Database and Expert Systems Applications - Volume 9261Tweets spread on social micro-blog bears some similarity to epidemic spread. Based on the findings from a user study on tweets' short-term retweeting characteristics, we extend the classic Susceptible-Infected-Susceptible SIS epidemic model for tweet's ...
Comments
Information & Contributors
Information
Published In
September 2022
185 pages
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than the author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from [email protected].
Publisher
Association for Computing Machinery
New York, NY, United States
Publication History
Published: 09 March 2022
Accepted: 01 October 2021
Revised: 01 September 2021
Received: 01 May 2021
Published in TSAS Volume 8, Issue 3
Check for updates
Author Tags
Qualifiers
- Research-article
- Refereed
Funding Sources
- Grant-in-Aid for Young Scientists
- Promotion of Science, Japan’s Ministry of Education, Culture, Sports, Science, and Technology (MEXT), and Strategic International Collaborative Research Program (SICORP)
- Japan Science and Technology Agency
Contributors
Other Metrics
Bibliometrics & Citations
Bibliometrics
Article Metrics
- 0Total Citations
- 859Total Downloads
- Downloads (Last 12 months)279
- Downloads (Last 6 weeks)43
Reflects downloads up to 23 Jan 2025
Other Metrics
Citations
View Options
View options
View or Download as a PDF file.
PDFeReader
View online with eReader.
eReaderHTML Format
View this article in HTML Format.
HTML FormatLogin options
Check if you have access through your login credentials or your institution to get full access on this article.
Sign in