GA-LightGBM模型及其在车辆保险需求预测中应用

Genetic Algorithm-LightGBM Model for Vehicle Insurance Demand Forecasting

为了提高LightGBM模型在车辆保险需求预测的准确率,引入遗传算法来优化LightGBM模型的参数,提出了一个GA-LightGBM模型。该模型主要分为3步:(1)对数据集进行预处理,包括特征描述性分析、去除无效值、分类特征数值化以及数值特征标准化;(2)使用遗传算法快速随机的全局搜索能力优化LightGBM模型参数;(3)根据最优参数组合训练LightGBM模型,并将最优模型应用于车辆保险需求预测中。实验结果表明在车辆保险需求预测方面,采用GA-LightGBM模型在均方根误差、召回率、F1值和AUC值相较于网格搜索法以及贝叶斯搜索法均有提升,模型性能均优于随机森林、GBDT、Bagging和Adaboost,可为保险公司商业决策提供参考。

In order to improve the accuracy of the LightGBM model for vehicle insurance demand forecasting,a GA-LightGBM model is proposed by introducing a genetic algorithm to optimise the hyperparameters of the LightGBM model.The model is divided into three main steps:(1)pre-processing of the dataset,including descriptive analysis of the features,removal of invalid values,numericalisation of classifcation features and normalisation of numerical features;(2)optimisation of the LightGBM model hyper parameters using the fast and random global search capability of the genetic algorithm;(3)fnally training the LightGBM model according to the optimal hyper parameter combination and applying the optimal model to forecast vehicle insurance demand.The experimental results show that the GA-LightGBM model has improved the root mean square error,recall,F1and AUC values compared with the grid search method and Bayesian search method in vehicle insurance demand forecasting,and the model outperforms Random Forest,GBDT,Bagging and Adaboost,which can provide a reference for making decisions on insurance companies’businesses.