Research ArticlePUBLIC HEALTH

Fine-grained dengue forecasting using telephone triage services

See allHide authors and affiliations

Science Advances  08 Jul 2016:
Vol. 2, no. 7, e1501215
DOI: 10.1126/sciadv.1501215
  • Fig. 1 Trends in call volume and suspected dengue cases measured during 2012 and 2013.

    (A) Time series of calls (red), suspected dengue cases (black), and awareness campaigns (green points). Scale normalized by dividing by individual maximum values. The x-axis label is in week of the year. (B) Density map of calls across towns in Lahore. (C) Density map of cases across towns in Lahore. The lightest shade represents the least number, and the darkest shade represents the highest number. The legend is normalized by the maximum value. Lat, latitude; long, longitude.

  • Fig. 2 Town-wise predictions of log-suspected cases from the ensemble model based on calls and weather data.

    Suspected dengue cases (black) and predictions from the model (red).

  • Fig. 3 Town-wise predictions of log-suspected cases from the ensemble model based on calls, cases, and weather data.

    Suspected dengue cases (black) and predictions from the model (red).

  • Fig. 4 Punjab Health Hotline Reporting System.

    (A) Interface used by operators to lodge complaints. (B) Interface used by officials to view complaints. (C) Types of complaints being lodged in the system. (D) Front-end interface of our dengue cases prediction system.

  • Table 1 Random forest importance weights for parameters of the model trained over the total year and season (July to November).
    ParameterAverage IncNodePurity
    TotalSeason
    Calls424.34172.08
    Awareness274.28158.41
    Rainfall123.1450.89
    Humidity287.4081.65
    Temperature349.38137.61

Supplementary Materials

  • Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/2/7/e1501215/DC1

    fig. S1. Cross-correlation between suspected incidences reported at hospitals and calls received at the health hotline in Lahore for the year 2012.

    fig. S2. Town-wise predictions of 3-week log-suspected incidence forecast from ensemble model based on calls and weather data.

    fig. S3. Town-wise predictions of 3-week log-suspected incidence forecast from ensemble model based on calls, cases, and weather data.

    fig. S4. Town-wise predictions of 2-week log-suspected incidence forecast from generalized linear model based on calls and weather data.

    fig. S5. Town-wise predictions of 2-week log-suspected incidence forecast from location-dependent ensemble model based on weather data.

    table S1. RMSE values between predicted and actual number of log-suspected cases for various models trained on coarse-grained city-level data set.

  • Supplementary Materials

    This PDF file includes:

    • fig. S1. Cross-correlation between suspected incidences reported at hospitals and calls received at the health hotline in Lahore for the year 2012.
    • fig. S2. Town-wise predictions of 3-week log-suspected incidence forecast from ensemble model based on calls and weather data.
    • fig. S3. Town-wise predictions of 3-week log-suspected incidence forecast from ensemble model based on calls, cases, and weather data.
    • fig. S4. Town-wise predictions of 2-week log-suspected incidence forecast from generalized linear model based on calls and weather data.
    • fig. S5. Town-wise predictions of 2-week log-suspected incidence forecast from location-dependent ensemble model based on weather data.
    • table S1. RMSE values between predicted and actual number of log-suspected cases for various models trained on coarse-grained city-level data set.

    Download PDF

    Files in this Data Supplement: