| In the information age,with the improvement of data acquisition capabilities and the advancement of storage technology,how to obtain valuable information from data has become a hot research topic in many industries.However,in the real world,there is a problem of class imbalance in many data.Traditional machine learning algorithms have poor performance in classifying imbalanced data,while ensemble learning has become a research hotspot in imbalanced data classification due to its excellent generalization ability,which reduces the impact of imbalanced data on a single classifier.This thesis proposes two imbalanced data classification methods from the perspectives of resampling and multi-objective optimization ensemble,and applies them to the auxiliary diagnosis of breast cancer.The main work of this thesis is as follows:(1)In order to pay more attention to overlapping samples in the process of oversampling and reduce the overfitting problem of the classifier for overlapping minority samples,an imbalanced data ensemble classification method based on local sampling was proposed.Firstly,a weighted enumeration sampling strategy based on the difficulty coefficient is introduced,which marks the difficulty level of overlapping samples using the difficulty coefficient,and iteratively generates particles with different spatial distributions to guide the construction of diverse sample subsets.Then,the proposed data processing strategy based on Gaussian perturbation is used to perturb the minority samples repeatedly selected in the sample subset,so as to reduce the classifier’s overfitting of overlapping minority samples repeatedly selected.The experimental results on multiple real imbalanced datasets show that BLS-IE can improve the classification performance of the classifier on imbalanced data at the data level.(2)In order to solve the problem of undersampling methods easily losing important information and the weak stability of a single classifier resulting in suboptimal multi-objective optimization ensemble,a imbalanced data classification method based on multi-objective optimization layer ensemble(MOILE)is proposed.Firstly,based on the ideas in BLS-IE,an improved density based undersampling method is improved by combining sample distribution information,which reduces the risk of losing important information for most class samples while increasing the diversity of sample subsets.Then,combined with a hierarchical ensemble strategy based on multi-objective optimization,the heterogeneous basis classifiers trained from the sample subset are integrated as sub ensemble classifiers.The multi-objective optimization algorithm is used to optimize precision and recall,and multiple sub ensemble classifiers are selected for non-linear integration.Experiments were conducted on a large number of imbalanced datasets,and the effectiveness of this algorithm was comprehensively and systematically demonstrated by comparing it with existing mainstream imbalanced algorithms.(3)Design and implement an auxiliary diagnosis system for breast cancer based on multiobjective optimization and layer ensemble method.Design a human-machine interaction interface using Vue technology,implement the system’s business logic and core algorithms using Java and Python,and store the database using My SQL.The prototype system includes user login module,data management module and breast cancer auxiliary diagnosis module based on local sampling and layer ensemble method.The system can effectively realize the auxiliary diagnosis of breast cancer. |
