Data mining can make early warning system to predict dengue possible

By Jyoti Singh 

New Delhi: A new study has shown that an early warning system may be developed to predict dengue outbreaks by mining meteorological and epidemiological data for a given geographic area.

In the study, scientists at the Indian Institute of Remote Sensing, Dehradun, used data mining tools to analyze weather parameters before and during dengue outbreaks in Delhi. Using two different statistical approaches – multivariate regression and Naive Bayes approach – they attempted to estimate how weather conditions may influence intensify or lead to a more severe outbreak and which parameters are strongly correlated with this metric.

The results showed that locations or regions in the national capital – Central Delhi, Civil Lines, Rohini, South and West Delhi – have the highest odds of dengue occurrence. While studying transmission, researchers also located specific zones in the city that share a similar disease-spreading pattern. The proposed model has crystallized the relationship between weather and dengue outbreak. The study suggests that sudden and high rainfall accompanied with 30-35 degree Celsius temperature and high relative humidity makes conditions highly vulnerable for the spread of dengue.

Data for the study was collected from various sources. The data on monthly dengue cases was collected from the Government of NCT. For the geospatial study zone-wise distribution of dengue cases, data came from the National Vector Borne Disease Control Programme (NVBDCP). Information on monthly weather parameters – rainfall, relative humidity and temperature – was collected from the Statistical Abstract of Delhi and India Meteorological Department (IMD). The temperature, relative humidity and rainfall data of 10 automated weather stations of Delhi were collected from IMD.

“The study may be helpful in designing an early warning system that can help decision makers to take measures to prevent and control dengue outbreaks,” said the researchers.

“The availability of reliable and segregated data on disease prevalence was not available, which was a major limitation of the study. The researchers had to do away with modelling on certain parameters, weakening the models and reducing the accuracy of predictions,” said V R Raman, a public health expert not connected with the study.

Raman added that knowing the highest probable localities for disease outbreak becomes more important for ensuring adequate public health planning, resilience and response, than knowing the large areas or clusters where the outbreak probabilities are high. Demographic information is critical too. “Whether the data used for modelling was comprehensive and reliable is yet another critical question, given the poor state of affairs around disease databases in our country.”

The research team includes Shiva Reddy Koti, Nikita Agarwal, Sameer Saran and A. Senthil Kumar. The study has been published in journal Current Science.

(India Science Wire)