Weather integrated malaria prediction system using Bayesian structural time series model for northeast states of India
by Hariprasad Vavilala, Nikhila Yaladanda, Phani Krishna Kondeti, Rafiq Unissa, Rajasekhar Mopuri, Krushna Chandra Gouda, Kantha Rao Bhimala, Madhusudhan Rao Kadiri, Suryanaryana Murty Upadhyayula & Srinivasa Rao Mutheneni
Malaria is an endemic disease in India and targeted to eliminate by the year 2030. The present study is aimed at understanding the epidemiological patterns of malaria transmission dynamics in Assam and Arunachal Pradesh followed by the development of a malaria prediction model using monthly climate factors. A total of 144,055 cases in Assam during 2011–2018 and 42,970 cases in Arunachal Pradesh were reported during the 2011–2019 period observed, and Plasmodium falciparum (74.5%) was the most predominant parasite in Assam, whereas Plasmodium vivax (66%) in Arunachal Pradesh. Malaria transmission showed a strong seasonal variation where most of the cases were reported during the monsoon period (Assam, 51.9%, and Arunachal Pradesh, 53.6%). Similarly, the malaria incidence was highest in the male population in both states (Asam, 55.75%, and Arunachal Pradesh, 51.43%), and the disease risk is also higher among the > 15 years age group (Assam, 61.7%, and Arunachal Pradesh, 67.9%). To predict the malaria incidence, Bayesian structural time series (BSTS) and Seasonal Auto-Regressive Integrated Moving Average with eXogenous factors (SARIMAX) models were implemented. A statistically significant association between malaria cases and climate variables was observed. The most influencing climate factors are found to be maximum and mean temperature with a 6-month lag, and it showed a negative association with malaria incidence. The BSTS model has shown superior performance on the optimal auto-correlated dataset (OAD) which contains auto-correlated malaria cases, cross-correlated climate variables besides malaria cases in both Assam (RMSE, 0.106; MAE, 0.089; and SMAPE, 19.2%) and Arunachal Pradesh (RMSE, 0.128; MAE, 0.122; and SMAPE, 22.6%) than the SARIMAX model. The findings suggest that the predictive performance of the BSTS model is outperformed, and it may be helpful for ongoing intervention strategies by governmental and nongovernmental agencies in the northeast region to combat the disease effectively.