FluCast - Android Application to Forecast Flu

512 views

Published on

FluCast - Android Application to Forecast Flu

Published in: Data & Analytics
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
512
On SlideShare
0
From Embeds
0
Number of Embeds
9
Actions
Shares
0
Downloads
7
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide
  • “Hello Everyone, Today we are going to present our application for the Biosurvelliance Mobile App Competition.”
    “Our Mobile App – Flucast – is an app to assess risk and forecast the seasonal flu in real time for USA.”
    “We are the interns from Android Team One – from Data Science and Analytics Group”
  • Before going into the details of the Mobile App, let me - list the main reason for choosing Flu;

    Flu is an annually recurring disease characterized by the prevalence of outbreaks.
    In USA alone, the cost of influenza epidemics to the economy is $ 71-167 billion per year.
    Rate of flu transmission is high, so early detection of the epidemic will help in reducing the cost of expenditure for the Health agencies

    Keywords associated with Flu are observed in feeds of social media which is a real-time source for disease analysis.
    Various data sources like weather are correlated with flu cases which help in risk assessment and risk prediction of the disease.
  • The Apps - Primary Target Customers are Epidemiologists who are responsible solely to monitor the disease outbreaks.
    In addition to the Primary, The Secondary Target Customers would be Airport officials, School Directors, company executives.

    In order to better understand the customer’s need, we conducted interviews with Mary Lancaster, an epidemiologist at PNNL. With her feedbacks and suggestions – we have compiled the following list of customer needs -

    Tool or Technology to Help in decreasing the time to action which will help stop the outbreak of influenza.
    Enhanced collaboration between the analysts involved in the early decision making for flu.
    Simultaneous Access to the different data sources like weather, social media and ILI cases to study the correlation.
    Targeted news from flu related websites and also access to flu related social media banter.
    An Analysis tool that is mobile enough to change certain parameters that help in decision making.
  • App Flow for Flu Forecast
  • --In flu forecast feature, we predict seasonal flu based on the data from ILINET flu cases, which is the influenza-like-illness weekly report, and weather data consisting of precipitation, temperature and wind speed. In addition, we are trying to integrate the social media mentions containing flu related keywords.
    --The forecast time period is 3 weeks ahead of the CDC report on the ILINET cases.
    --In the prediction analysis, we would be using ARIMA Model. In the final risk forecast, data from weather and social media were included into the analysis.
    -- Forecasted estimate of flu cases with an error rate of 11,18 and 24% for three weeks of prediction
  • In addition, we provide 95% confidence interval for the number of flu cases.
  • Region level data is ILINET cases and State Level Cases are calculated
    Scheduled Automation of data retrieval for CDC cases and Google Flu Trends.
    Programmatically access weather and twitter data for analysis.
    Flask Python Web framework built for REST API to interface with Database which also suits the python data analysis framework that we have built
    Mongo Database to store time series data.
    Angularjs for the data model and Jquery Mobile for the interface design to build HTML5 App.
    Phonegap framework to convert HTML5 App to Android Native App. Helping us build the app for multiple platforms like Android, iOs and Windows phone with a single codebase.


  • App Flow for Data Sharing
  • --With interesting analysis scenario, the app provides an ability to share the information in the form of screenshots and tabular data form to other analyst to enable collaboration.
    -- The app provides an ability to create groups based on the needs of the analyst.
    --The conversation between different analyst through the phone’s email application, thus enabling the analyst not take extra effort to use our app.
    --The App will be a convenient platform for the analyst to communicate and collaborate for flu disease.
  • Ability to share every screen with a screenshot of the active window and simultaneously data in a tabular form for the analysts own analysis
    Without the app being a broker for communication by allowing the analysts to continue the communication over email
    Creating groups based on the contacts will enable the analyst to easily import the contacts from their phones
  • App flu for Flu Stream
  • Flu stream allows the app users to be updated with the flu related news through different websites from CDC, WHO and other flu related websites.
    Real time Twitter stream of data with geo-location of the tweet, with the tweets being filtered for flu related keywords.
    Allow the analyst to read the complete article for the flu in the mobile app.
    In addition, the ability to share the article on third party apps on the android phone.
  • - These Flu stream is updated every hour and categorized with time so users could keep track of the latest flu situation.
    - The sharing to third party apps features is implemented by the android in-built share function.
    Flu News and Tweets from the twitter are geotagged to identify the location of the news and help the analyst the threat region for the particular news item.
    Using elastic-search database to store the tweets, with its built in twitter river to store 1% of twitter feed data that is streamed real time to all the developers.
    Geo-Tagging of the flu news and tweets are done by tokenizing the summary of the article and twitter users location by Natural Language processors in Python. Twitter users profile location is found to be 80% accurate prediction of the tweet location. Geotagging is at the level of states.

  • Monitoring twitter for real-time stream of data, which is updated every day and monitored for any unpredictable changes in the flu patterns.
    Threshold value for Outbreak alert is calculated from the historical tweets data on a weekly basis with exponential smoothing model.
    Push notification to notify the analyst about any potential outbreaks in the region concerned with the flu mentions crossing the high threshold value set for the region or state.

  • Real time monitoring of twitter from the twitter real-time data stream allows to data analysis in real time enabling changes to the number of flu mentions to alert the analysts about any potential outbreak.
    Tweets are geo-tagged in real-time and outbreak is based on the threshold value and pushed to the user through Google Cloud Messaging
    Push Notification would contain related information about the region associated with the potential outbreak and link to tweets that are associated with the analysis
  • Columbia Prediction of Infectious Diseases : Influenza Forecasts (Web App);
    SickWeather (iOS App);
    FluSpotter ( Facebook App);
    Fludar ( Web App)

    By researching about similar competitors in the area which could help epidemiology experts; we observed a general lack of features -
    -- Their lack of mobility hinders users ability to use tools on the move.
    -- They don’t have features that enable sharing between individuals or in a group.
    -- They do not use the real-time data for risk assessment or notify outbreaks. Their data sources are very limited, either no weather or social media data.
    -- Many of these apps are just information visualization with no analysis involved in the final presentation.
  • Yixin Hu
    Columbia University

    Lasifu Ta
    University of Washington

    Sandeep Shantharam
    Indiana University Purdue University, Indianapolis
  • FluCast - Android Application to Forecast Flu

    1. 1. FluCast Risk assessment and Forecast of seasonal flu in real time YIXIN HU, SANDEEP SHANTHARAM AND LASIFU TA Data Sciences & Analytics Group, Computational & Statistical Analytics Division, National Security Directorate PNNL-SA-104651
    2. 2. FluCast - Mobile App for Risk and Forecast of Flu Why Flu? Infectious respiratory disease causing Seasonal Epidemic $71-167 billion per year spent on “Flu” in the U.S. Transmission rate between humans is high Keywords of flu symptoms are expressed on Social Media Rich and Varied data sources for disease analysis August 6, 2014 2
    3. 3. FluCast - Mobile App for Risk and Forecast of Flu Need for Mobile App Target Customers - Epidemiologists – primary - Transport / School / Company Officials – secondary Customer Needs - Decrease time to action - Enhance collaboration - Access to varied data sources - Targeted flu news - Mobile analytics app August 6, 2014 3
    4. 4. FluCast - Mobile App for Risk and Forecast of Flu Flu Forecast August 6, 2014 4
    5. 5. FluCast - Mobile App for Risk and Forecast of Flu Flu Forecast – Approach Prediction with data from historical flu cases, weather and social media. ARIMA Model for flu prediction with Time-Series data. Forecasting three weeks ahead of CDC. Estimated flu cases with accuracy about 90% August 6, 2014 5
    6. 6. FluCast - Mobile App for Risk and Forecast of Flu Flu Forecast – Approach August 6, 2014 6 H1N1
    7. 7. FluCast - Mobile App for Risk and Forecast of Flu Flu Forecast – Implementation Automated data retrieval from CDC and Programmatically access weather and twitter data from NOAA and Twitter Flask web framework for backend REST API with Mongo Database suited for time series data Jquery Mobile to build the HTML5 App PhoneGap framework for transformation August 6, 2014 7
    8. 8. FluCast - Mobile App for Risk and Forecast of Flu Flu Forecast – UX Design August 6, 2014 8
    9. 9. FluCast - Mobile App for Risk and Forecast of Flu Flu Forecast – UX Design August 6, 2014 9
    10. 10. FluCast - Mobile App for Risk and Forecast of Flu Data Sharing August 6, 2014 10
    11. 11. FluCast - Mobile App for Risk and Forecast of Flu Data Sharing – Approach Information Sharing and Collaboration between Epidemiologists Sharing data within a group of trusted individuals Communication between the analysts over E-mail Sharing public domain information to Social Network August 6, 2014 11
    12. 12. FluCast - Mobile App for Risk and Forecast of Flu Data Sharing – Implementation Sharing the screenshot of the active window and data in tabular form Information sharing through the phone’s email application Ability to create groups with the contacts on the analyst’s phone Sharing public level data to Social Networks like Facebook, Twitter August 6, 2014 12
    13. 13. FluCast - Mobile App for Risk and Forecast of Flu Data Sharing – UX Design August 6, 2014 13
    14. 14. FluCast - Mobile App for Risk and Forecast of Flu Flu Stream August 6, 2014 14
    15. 15. FluCast - Mobile App for Risk and Forecast of Flu Flu Stream – Approach News stream from RSS feeds of flu related websites like CDC and WHO Twitter real-time stream with geo-location for tweets related to Flu Stay updated with the news from the RSS and Twitter in real time Share the News article and Tweets to Social Network August 6, 2014 15
    16. 16. FluCast - Mobile App for Risk and Forecast of Flu Flu Stream – Implementation Flu Stream is updated every hour for both RSS and Twitter Stream Items in News stream and Twitter stream are geotagged to allow analysts identify the location associated Geotagging by the list of keywords based on states and usage of natural language processing to tokenize the keywords Sharing to all third party apps with the phones in-built share function August 6, 2014 16
    17. 17. FluCast - Mobile App for Risk and Forecast of Flu Flu Stream – UX Design August 6, 2014 17
    18. 18. FluCast - Mobile App for Risk and Forecast of Flu Flu Stream – UX Design August 6, 2014 18
    19. 19. FluCast - Mobile App for Risk and Forecast of Flu Outbreak Alert August 6, 2014 19
    20. 20. FluCast - Mobile App for Risk and Forecast of Flu Outbreak Alert – Approach Real-time monitoring of Twitter stream for the flu symptom mentions Exponential Smoothing Model used in calculating the threshold value Push Notification to the analyst, for possible outbreaks on a daily basis August 6, 2014 20
    21. 21. FluCast - Mobile App for Risk and Forecast of Flu Outbreak Alert – Approach August 6, 2014 21 H1N1
    22. 22. FluCast - Mobile App for Risk and Forecast of Flu Outbreak Alert – Implementation Live data stream from twitter allows for real-time analysis Tweets are automatically geotagged as the data is streamed Push notification to the analyst based on the twitter mentions being higher than threshold value for high risk Push Notification to the device with related information gathered from twitter for Outbreak Conditions August 6, 2014 22
    23. 23. FluCast - Mobile App for Risk and Forecast of Flu Outbreak Alert – UX Design August 6, 2014 23
    24. 24. FluCast - Mobile App for Risk and Forecast of Flu Competition August 6, 2014 24
    25. 25. FluCast - Mobile App for Risk and Forecast of Flu Benefit August 6, 2014 25
    26. 26. FluCast - Mobile App for Risk and Forecast of Flu Future Direction July 22, 2014 26 Option for robust prediction models Offline access to the app Subtyping Flu Data source from multiple social networks Better Twitter Data Analysis Analyst reporting
    27. 27. FluCast - Mobile App for Risk and Forecast of Flu Acknowledgement August 6, 2014 27 Technical Staff Harisson, Joshua J Dowling, Chase P Henry, Michael J Franklin, Lyndsey R Lancaster, Mary Managerial Staff Spence, Christine N Noonan, Christine F Cowley, Wendy E Corley, Courtney D Didier, Brett T
    28. 28. FluCast - Mobile App for Risk and Forecast of Flu Yixin Hu – Data Analyst Columbia University Android Team One Lasifu Ta – UX Designer/ Front-end University of Washington Sandeep Shantharam – Developer Indiana University Purdue University, Indianapolis Thank you! August 6, 2014 28

    ×