Improvement Of TrAdaBoost Algorithm In Structured Data Outliers

Transfer learning has attracted much attention in recent years because it can solve the problems caused by traditional machine learning algorithms to a certain extent due to the lack of labeled target domain data.TrAdaBoost is an instance-based transfer learning method proposed by Dai et al.In 2007.In recent years,the algorithm has been used in image classification,text mining,and other problems,but it has fewer applications in structured data.In this paper,the algorithm is applied to structured data,and the problem of the surge of errors on the structured data set Mushroom dataset used in the original text is found.Based on this problem,an improved algorithm is proposed.The improved algorithm is compared with the SVM algorithm using only the target domain data,the SVM?t algorithm using both the target domain data and the auxiliary domain data,and the TrAdaBoost algorithm in different data sets.It is found that our improved algorithm has obvious improvement on each evaluation index of each data set compared with the other three algorithms.We believe that our improved algorithm can be applied to instance-based transfer learning problems in structured data,and it is improved compared to TrAdaBoost algorithm.Therefore,we believe that this algorithm can be applied to structured data,and it is an effective transfer learning method in structured data,which can achieve good results.