Optimized soft voting ensemble model for predicting net blotch disease in spring barley with meteorological data in Turkey

Early detection of plant diseases helps reduce the use of pesticides in agricultural production and increases productivity. Methods based on meteorological data are successful in predicting plant diseases and weather data are effective in disease development. In this study, a unique ensemble model was proposed to detect the presence of net blotch disease in spring barley in Turkey. Nine different meteorological data collected between 2021-2024 and disease occurrence status were used as data for classification. The performance of 12 different classifiers was evaluated on the collected data. Among them, Multi-Layer Perceptron (MLP) achieved the highest accuracy with 89.6%. However, while Machine Learning approaches reduce misclassification costs for unbalanced datasets, they struggle to achieve high prediction accuracy. Therefore, the NearMiss-2 method was applied to resolve the class imbalance and 5 different voting-based ensemble approaches including 2,3,4,5 classifiers were proposed and tested to improve the diagnostic accuracy. The results showed that the combination of 3 classifiers consisting of ET, RF and MLP provided an improved classification accuracy of 91.9% and this ensemble method was effective in detecting net blotch disease. For the same model, the precision, recall, f-score and AUC values were 89.5%, 94.4%, 91.9% and 0.945, respectively. The combination of the three classifiers using the majority rule consistently outperformed the classifiers working alone, showing that it facilitates robust decision making by striking a balance between diversity and complexity. The high results obtained showed an improvement in the detection of net blotch in barley compared to previous studies in the literature. It is demonstrated that the model can be used to detect diseases and thus plan pest control methods and pesticide use in crops.