Transfer learning enables predictions in soil-borne diseases
Soil Ecology Letters(2024)
摘要
Inhibiting the occurrence of soil-borne diseases is considered as the most favorable approach for promoting sustainable agricultural development. Constructing soil disease prediction models can serve precision agriculture. However, the analysis results of the metaframework often contradict each other, causing inconsistency in the important features of machine learning results. Therefore, it is necessary to compare the classification accuracy of various machine learning models and further optimize the features of the models to enhance their classification accuracy. Here, we conducted a comparison of eight common machine learning algorithms (XGBoost, CatBoost, Decision Tree, LGBM, Naïve Byes, Perceptron, Logistic, and Random Forest) at the levels of family, genus, and class. The important features of the model were extracted based on the differences in model accuracy and important features, followed by an interpretable analysis of these important features using feature importance. Subsequently, the data underwent resampling using the SMOTE algorithm, and the results show that the SMOTE-Transformer model performs well, surpassing the training results of the voting and stacking strategies, with an accuracy reaching 90
更多查看译文
关键词
soil disease,feature importance,heterogeneous integration strategy,transfer learning
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要