Quality classification of stored wheat based on evidence reasoning rule and stacking ensemble learning

Abstract

Classification of stored wheat quality is extremely important for reducing wheat storage losses and ensuring food security. However, the limitations of conventional single models and the uncertainties in the decision-making process have brought great challenges to the quality classification of stored wheat. To solve this problem, an accurate classification model for stored wheat quality based on Evidence Reasoning rule and Stacking ensemble learning (ER-Stacking) is proposed. Firstly, the base learners of the ensemble model are selected by the fusion measurement method of diversity and classification performance. Then, the importance weight and reliability factor of the evidence are obtained by the Differential Evolution (DE) algorithm and probability distance similarity, respectively. Finally, the ER rule is used to fuse the evidences optimized by weight and reliability to complete the identification and classification of stored wheat quality. In order to verify the validity of the method, experiments are conducted using the structured data on physiological and biochemical indicators of stored wheat. The experimental results show that the ER-Stacking ensemble model achieves 88.1%, 88.05%, 89.31% and 88.4% in the accuracy, precision, recall and f1-score, respectively, whose classification performance is significantly higher than that of the conventional single models. Compared with the models using other integration methods or different combination strategies of base learners, the proposed model also has obvious advantages, which can effectively improve the accuracy and reliability of the classification results of stored wheat quality.