Proximity and Risk with Covid-infected Places for Mobile Users:
A Big Data Machine Learning Approach

Guanyao Li, Tianlang He, Wai Lun Tsui, Ki Kit Lai, Gary Chan

Department of Computer Science and Engineering

The Hong Kong University of Science and Technology

{gliaw, theaf, ptsuiwl, gchan}

Covid-19, especially its Omicron variant, is spreading fast globally taking toll on many lives. The infection is often caused by exposure to the virus for sustained period of time due to close distancing with the contaminated places, the so-called proximity risk. Proximity risk is related to not only exposure duration and physical distance to the Covid-infected places, the so-called “incident places,” but also the location noise and the number and nature of the venues (restaurants or places with large number of confirmed cases are often of higher risk). We have designed and developed a technology and an app called CovidInArea 疫地而處 to provide in private timely information to mobile users on the infected places in their neighborhood, and show the proximity risk with these places over time (in Apple App Store, search for either covidinarea or 疫地而處, while in Google Play Store, search for covidinarea or 地而處 without 疫). Armed with such knowledge, mobile users may then take appropriate actions to plan routes, manage one’s health, and keep safe distancing from these places.

To better inform citizens, The Government of HKSAR currently makes known to the general public a dashboard showing buildings with case resided/visited in the past 14 days. However, these data are presented as a static webpage, which requires mobile users to know their locations without real-time GPS support. Some apps leverage the data by asking users to input locations in order to check manually if they are in proximity to the incident places. They are mostly not automatic, and offer no assessment or visualization of proximity risk over time for users to take appropriate proactive measures against infection.
A heatmap of infected places in Hong Kong.
Using the open Data in Coronavirus Disease (COVID-19) provided by the Department of Health accessible through (“List of buildings visited by cases tested positive for SARS-CoV-2 virus in the past 14 days”), we have designed and developed a novel, private and automated mobile-friendly app CovidInArea which integrates the data and visualizes them as an easily accessible heatmap pinpointing the incident places as hotspots. Mobile users may browse the heatmap to understand where the places with infected cases are, and make informed decision on whether they should keep safe distancing from them. They may optionally turn on their GPS to check automatically and locally the proximity risk in real time. Our mobile computing system makes use of the following approaches:
  • Data mining to assess proximity risk [1, 2]: We integrate the open and trustworthy data sources provided by Department of Health and present it as an informative, mobile-friendly and user-accessible heatmap for browsing, visualization and understanding at a quick glance where the incident places are. The heatmap is updated continuously according to government release. If the user turns on GPS, the heatmap can be used to automatically pinpoint one’s current location with respect to these areas, and the app alerts the user if such buildings are in proximity. The number of incident places in proximity is displayed in real time. Furthermore, the overall proximity risk is indicated as a colored polygon of a radar chart indicating the quantity level of places, contact duration and proximity level. Based on its color, the user may take timely and appropriate actions to manage health, plan routes and keep proper distancing to reduce infection risk:

    Red: Overall high sustained proximity level to significant number of incident places. Recommended to avoid high-risk places, manage health and take voluntary testing if necessary.

    Yellow: Medium risk. Be cautious. Plan safe paths to reduce risk of infection.

    Green: Low risk. Stay vigilant.

  • Machine learning for power-conserving GPS sampling [3]:To conserve battery consumption due to GPS, we have devised and implemented a light-weight machine learning approach to adaptively adjust GPS sampling according to user mobility level. This greatly reduces the GPS sampling rate, and hence power consumption, by more than 90%.
  • The heatmap of places with Covid cases.
    The places with Covid cases in one's neighborhood.
    Number of places with Covid cases within 100m and overall proximity risk.
    User privacy is our primary design consideration. Our system is fully distributed without maintaining any server to collect user information (Privacy Policy). CovidInArea achieves the highest user privacy with the following measures:
  • No user registration: Users do not need to register in order to use the app.
  • No personal information collection beyond GPS location: The app respects data minimization in privacy by design. It does not ask for any sensitive information beyond the optional GPS location. It uses no personal and phone data, such as phone number, contact list, photo gallery, etc.
  • No any upload of user data: Only consolidated publicly available data from the government are downloaded to the users. No any user data including location leaves the phone to our system. All computations are carried out with results presented locally in one’s phone.
  • No location storage at any time:Users may opt to turn on their GPS at any time for live proximity checking and risk assessment. The GPS location, once consumed, is immediately discarded without storage at any time. No any personal identifiable information is collected and stored. The app keeps no record of proximity venues, and any data older than a certain past period of time (e.g., 3 days) are purged.
  • The app CovidInArea is currently available in Google Play Store and Apple App Store in Hong Kong. Users may install it using their phones by searching for “CovidInArea” in the stores, visiting the link, or scanning or clicking the following QR code:
    Press coverage
  • 中国新闻社 - 港科大团队自主研发防疫App 科研人员:将充分保障用户隐私
  • 广东省港澳办 – 港科大推首个本地院校研发的防疫App
  • RTHK32 – An interview by RTHK32 (video time 7:50-15:55)
  • SCMP – Hong Kong research team creates hi-tech app designed to assess exposure to coronavirus
  • The Standard – Find Covid cases around you with map app
  • 中新網 – HKUST launches app for COVID-19 prevention and control
  • 香港01 – 科大研發手機程式 實時顯示確診個案分佈 分析用戶感染風險
  • 星島日報 – 科大研發疫廈分布APP可顯示3日到訪區域風險
  • 星島日報 - 科大推首個本地院校研發防疫App 視像式顯示個案大廈分布
  • 香港經濟日報 – 【確診大廈】科大研發顯示個案大廈分布地圖App 3種顏色顯示高中低風險
  • Topick – 「疫地而處」App 顯示個案大廈分布
  • 東網 – 科大團隊首研手機應用程式 助市民了解疫廈分布及感染風險
  • Line Today – 科大推首個本地院校研發防疫App 視像式顯示個案大廈分布
  • 頭條日報 – 科大推首個本地院校研發防疫App-視像式顯示個案大廈分布
  • 雅虎(香港) – 科大推首個本地院校研發防疫App 視像式顯示個案大廈分布
  • 晴報網 – 科大「疫地而處」APP實時標示疫廈密度
  • e-zone – 科大研發防疫 App「疫地而處」 顯示高危地點實時計算用家感染風險
  • am730網 – 疫情|科大推程式「疫地而處」 以GPS顯示確診大廈及風險
  • Line Today – 疫情|科大推程式「疫地而處」 以GPS顯示確診大廈及風險
  • 文匯報 – 科大推首個本地院校研發防疫App 視像顯示個案分布
  • 點新聞 – 科大推首個本地院校研發防疫App 視像顯示個案分布
  • 文匯網 – 科大「疫地而處」助規劃安全路線 手機App顯示個案大廈分布 不索取個人資料保私隱
  • 大公網 – 科大研發手機APP 顯示鄰近疫廈資訊
  • 香港商報網 – 科大推防疫App「疫地而處」 視像顯示個案大廈分佈
  • 巴士的報 – 科大推首個本地院校研發防疫App 視像式顯示個案大厦分布
  • 紫荊雜誌 – 科大推首個本地院校研發防疫App 以熱力圖形式顯示個案密度
  • 橙新聞 – 科大推防疫App「疫地而處」 個案大廈分布一目了然
  • 中國評論月刊網絡版 – 港科大推首個本地院校研發防疫App
  • 中新網 – HKUST launches app for COVID-19 prevention and control
  • HKUST - Privacy-Preserving App Maps Hotspots with COVID-19 Cases and Shows Proximity Risk
  • Reference
    [1] G. Li, S. Hu, S. Zhong, W. Tsui and S.-H. Chan, "vContact: Private WiFi-based IoT Contact Tracing with Virus Lifespan,'' IEEE Internet of Things Journal, Vol. 9, pp. 3465-3480, March 2022. [paper]
    [2] G. Li, C.-C. Hung, M.Liu, L. Pan, W.-C. Peng and S.-H. Gary Chan, "Spatial-Temporal Similarity for Trajectories with Location Noise and Sporadic Sampling," in Proceedings of 37th IEEE International Conference on Data Engineering (ICDE), Chania, Crete, Greece, pp. 1224-1235, 19 - 22 April, 2021. [paper]
    [3] T. He, J. Tan, W. Zhuo, M. Printz and S.-H. Chan, "Tackling Multipath and Biased Training Data for IMU-Assisted BLE Proximity Detection," in Proceedings of IEEE International Conference on Computer Communications (IEEE INFOCOM), Virtual Conference, IEEE, 2-5 May, 2022. [paper]