Prediction of monthly malaria outbreaks in districts of Odisha, India with meteorological parameters using statistical and artificial neural network techniques

Abstract

Malaria is a vector-borne disease spread by female Anopheles mosquitoes. This study provides the relationship between malaria and meteorological parameters over 10 districts in Odisha for the period 2012-2016. The complete life cycle of Plasmodium is dependent primarily on meteorological variables like rainfall, temperature, humidity. Rainfall increases the survival chances of mosquitoes by providing a habitat for the different development stages of mosquito larvae. Temperature and humidity affect the survival of Plasmodium and mosquitoes. Malaria cases peak in the monsoon season and decrease thereafter. The malaria cases have almost doubled over Odisha in 2014-2016 in comparison to 2012-2013. Minimum temperature (Tmin), Rainfall, and RH at noon show a significant maximum positive correlation with the malaria cases while the diurnal variations of temperature (DTR) and relative humidity are negatively correlated with the malaria cases. Almost all the peak occurrences of malaria are associated with the Tmin >20^o C range. DTR of 6-8^o C is associated with all of the peak malaria cases. The combination of all these meteorological variables decides the transmission of malaria at any place condition on the presence of Plasmodium in the vector mosquitoes.  The malaria  forecast models for different districts of Odisha are prepared using the relationship between meteorological parameters and malaria occurrence. The simple multiple linear regression and Artificial Neural Network (ANN) methods are applied for this purpose. The performance of ANN method is quite well compared to the multiple linear regression for almost all times. The RMSE range for Angul, Kandhamal, Mayurbhanj and Keonjhar for ANN methods is almost half compared to the Multiple linear regression methods. The lowest and highest RMSE in the ANN method is 152 in Keonjhar and 268 in Mayurbhanj, while the multiple linear regression method, is 339 for Ganjam and 776 in Mayurbhanj.  Also, the R-Square value is improved in ANN method compared to the Multiple linear regression methods.  Plasmodium