基于XGBoost的山核桃干腐病多分类预测模型研究

Research on Multi-Classification Prediction Model of Canker Carya Cathayensis Based on XGBoost Algorithm

【目的】通过分析影响山核桃短期内干腐病发病程度的因素,对比不同机器学习算法,得到发病程度预测效果较好的模型,为科学、绿色防治工作提供思路。【方法】采用机器学习算法XGBoost、逻辑回归和BP神经网络作为构建预测模型的基础,以五折交叉验证方法验证模型。【结果】对山核桃干腐病发病程度的影响因素排名,由高到低分别是候平均温度、候平均湿度、病斑数目、候降水量;集成机器学习算法XGBoost构建的预测模型各评价指标都高于逻辑回归和BP神经网络;集成机器学习算法XGBoost在山核桃干腐病发病程度多分类预测问题上得到的效果优于传统机器学习算法。

[Objective]By analysing the factors affecting the incidence degree of Carya cathayensis in canker in the short term,and comparing different machine learning algorithms to get the model with better prediction effect of incidence degree,we can provide ideas for scientific and green control work.[Method]The machine learning algorithms XGBoost,logistic regression and BP neural network were used as the basis for constructing the prediction model,and the model was verified by the five-fold crossvalidation method.[Result]Factors affecting the incidence degree of canker Carya cathayensis disease were ranked,from high to low,as pentad average temperature,pentad average humidity,number of spots,and pentad precipitation;each evaluation index of the prediction model constructed by the integrated machine learning algorithm XGBoost was higher than that of logistic regression and BP neural network;the integrated machine learning algorithm XGBoost obtained better results than the traditional machine learning algorithm in the multi-classification prediction problem of the incidence degree of canker Carya cathayensis disease better results than traditional machine learning algorithms.