An Improved Robust AdaBoost Regression Model Based on DBSCAN Algorithm
The traditional AdaBoost.R2 algorithm transforms the regression to a binary classification via AdaBoost algorithm and performs well.However,this algorithm continuously enlarges the weights of outliers during the iterations,then it is sensitive to outliers and lacks of robustness.This paper gives an improved AdaBoost regression model based on DBSCAN clustering algorithm,named as AdaBoost.DBSCAN,which identifies outliers in observations through DBSCAN,and different weight control strategies are adopted to ensure that the weight of outliers cannot increase exponentially in the itera-tive process,and sample information can be preserved to a large extent.Simulations and applications show that this new method outperforms AdaBoost.R2,AdaBoost.RT and AdaBoost.RS algorithms un-der different proportions of outliers.
NETL
NSTL
万方数据
