A Comparative Study of Ensemble Classifiers for Paddy Blast Disease Prediction Model

Paddy blast has become most epidemic disease in many rice growing countries. Various statistical methods have been used for the prediction of paddy blast but previously used methods failed in predicting diseases with good accuracy. However the need to develop new model that considers both weather factors and non weather  data called blast disease data that influences paddy disease to grow. Given this point we developed ensemble classifer based paddy disease prediction model taking weather data from January 2013 to December 2019 from Agricultural and Horticulture Research Station Kathalgere Davangere District. For the predictive model we collected 7 kinds of weather data and 7 kinds of disease related data that includes Minimum Temperature, Maximum Temperature, Temperautre Difference,Relative Humidity, Stages of Paddy Cultivation, Varities of seeds, Season of cropping and so on. It is observed and analyzed that Minimum Temperature, Humidity and Rainfall has huge correlation with occurrence of disease. Since some of the variables are non numeric to convert them to numeric data one hot encoding approach is followed and to improve efficiency of ensemble classifiers  4 different filter based features selection methods are used such as Pearson’s correlation, Mutual information, ANNOVA F Value, Chi Square. Three different ensemble classifiers are used as predictive models and classifiers are compared it is observed that Bagging ensemble technique has achieved  accuracy of 98% compared to Adaboost of 97% and Voting classifier of 88%. Other classification metrics are used evaluate different classifiers like precision, recall, F1 Score, ROC and precision recall score. Our proposed ensemble classifers for paddy blast disease prediction has achieved high precision and high recall but when the solutions of model are closely looked bagging classifier is better compared to other ensemble classifers that are proposed in predicting paddy blast disease