Application of machine learning models for risk estimation and risk prediction of classical swine fever in Assam, India.

ABSTRACT

The present study is aimed to develop an early warning system of Classical swine fever (CSF) disease by applying machine learning models and to study the climate-disease relationship with respect to the spatial occurrence and outbreaks of the disease in the north-eastern state of Assam, India. The disease incidence data from the year 2005 to 2021 was used. The linear discriminant analysis (LDA) revealed that significant environmental and remote sensing risk factors like air temperature, enhanced vegetation index, land surface temperature, potential evaporation rate and wind speed were significantly contributing to CSF incidences in Assam. Furthermore, the climate-based disease modelling was applied to relevant ecological and environmental risk factors determined using LDA and risk maps were generated. The western and eastern regions of the state were predicted to be at high risk of CSF with presence of significant hotspots. For the districts that are significantly clustered, the Basic reproduction number (R0) was calculated after the predicted results were superimposed onto the risk maps. The R0 value ranged from 1.04 to 2.07, implying that the eastern and western regions of Assam are more susceptible to CSF. Machine learning models were implemented using R statistical software version 3.1.3. The random forest, classification tree analysis and gradient boosting machine were found to be the best-fitted models for the study group. The models' performance was measured using the Receiving Operating Characteristic (ROC) curve, Cohen's Kappa, True Skill Statistics, Area Under ROC Curve, ACCURACY, ERROR RATE, F1 SCORE, and Logistic Loss. As a part of the suggested study, these models will help us to understand the disease transmission dynamics, risk factors and spatio-temporal pattern of spread and evaluate the efficacy of control measures to battle the economic losses caused by CSF outbreaks. Supplementary Information The online version contains supplementary material available at 10.1007/s13337-023-00847-6.