Prediction of potato late blight disease incidence based on weather variables using statistical and machine learning models: A case study from West Bengal

Abstract

Late blight is one of the most devastating diseases on potato the world over, including West Bengal, India. The economic and yield losses from outbreaks of potato late blight can be huge. In this article, application of statistical models such as autoregressive integrated moving average (ARIMA), autoregressive integrated moving average with exogenous variables (ARIMAX) in combination with machine learning models such as, neural network auto regression (NNAR), support vector regression (SVR) and classification and regression tree (CART) have been explored to predict the percentage disease index (PDI) of potato late blight in the northern part of West Bengal. Models were developed to predict PDI at 3- and 7-days interval using the weather variables viz., rainfall, maximum and minimum temperature, maximum and minimum relative humidity, and dew point temperature. Among the developed models, CART to predict PDI at 7 days interval was found to be the best fitted model on the basis of least RMSE, MAE and MAPE. The results of decision tree (CART) model showed that dew point temperature had a significant effect on PDI at 7 days interval and the incidence of potato late blight was high when dew point temperature was greater than 12 0C in the preceding week.