Risk assessment of corn borer based on feature optimization and weighted spatial clustering: a case study in Shandong...

As a typical pest affecting corn yield and safety, corn borer causes serious economic losses worldwide. Climate warming has intensified the occurrence of pest outbreaks in recent years, but the associated risk has not been precisely assessed or understood. To address this gap, this paper took Shandong Province, China as a case study, and constructed a feature optimization model for the class imbalance problem and a novel risk assessment method to quantify the temporal and spatial distribution of corn borer occurrence risk. Addressing the prevalent issue of class imbalance in pest datasets, a feature optimization model using Borderline-SMOTE to improve the Genetic Algorithm-Random Forest (GA-RF) was constructed, combined with Pearson correlation coefficient to jointly obtain a subset of features that affect corn borer. Subsequently, given the limitations of traditional risk assessment models that easily lose spatial information, by introducing the idea of weighted clustering algorithm, a novel machine learning model was proposed to assess the risk of agricultural pests and diseases. Finally, integrating natural disaster risk theory, this paper achieved an assessment and zoning of corn borer risk in the study area based on hazard, sensitivity, disaster prevention and mitigation capacity, and comprehensive states. The results indicated that compared with the original RF model, the improved feature optimization model achieves increases of 18.64%, 11.12%, and 11.21% in OOB_score, Accuracy, and F1_score, respectively, and outperforms eight other benchmark models. In terms of clustering performance, the weighted K-means clustering algorithm achieves higher Silhouette coefficient by 0.0138 and 0.1885 compared with the weighted agglomerative hierarchical clustering algorithm (weighted AHC) and weighted DBSCAN, respectively, the Calinski-Harabasz index is higher by 3.8017 and 22.4039, and the Davies-Bouldin index is lower by 0.1006 and 0.4889, demonstrating superior clustering results. The spatial zoning results closely align with actual conditions. The risk of corn borer occurrence was concentrated in the southwest and northern areas of Shandong Province, while the risk in the central and southeast areas was relatively low. This research provides a novel approach to agricultural disaster risk assessment and the obtained results can serve as decision support for corn borer prevention and control in Shandong Province.