Research On Ensemble Learning Method Based On Dual Evolutionary Architecture

The rapid development of information technology has endowed the data with massive,complex,and fast-changing characteristics.How to mine the intrinsic laws from the massive data to support decision-making has received widespread attention.Ensemble learning generates different base classifiers in a serial or parallel manner and combines the prediction results to make more informed decisions.This combination can effectively solve the shortcomings of a single individual,which improves the construction efficiency and generalization performance of the ensemble classifier.However,this mechanism cannot fully play the advantages of interaction and cooperation among base classifiers.It cannot simultaneously optimize the base classifier’s morphological structure and the ensemble classifier’s combined structure.The difficulty of ensemble learning lies in how to design an appropriate ensemble,such as the size of the ensemble,the selection of ensemble members,the setting of hyperparameters,etc.,which all lead to the performance of the ensemble model primarily relying on human design decisions.In this paper,evolutionary algorithms are used for base classifier generation and ensemble classifier construction,and the relevant theories and techniques are intensely studied.On this basis,the cooperative optimization between two levels of evolution is achieved,and the major parts accomplished are as follows:(1)Decision tree evolutionary search mechanism based on semantic structure and behavioral diversity.A decision tree is a tree model in which the middle node is a function for judging feature values,while the leaf nodes are the results of a series of function operations.In this paper,decision trees are used as the base classifier of the dual evolutionary ensemble and optimized using evolutionary algorithms.The diversity of the decision tree is increased from two perspectives of semantic structure and classification behavior,and the multi-objective fitness function is used to adjust the classification performance and complexity of the decision tree.The splitting and fusion operators achieve different granularity optimization of the decision tree structure and guide the decision tree to learn combinatorial knowledge in the evolutionary process,thus improving the generalization ability of the ensemble model.(2)Dual evolution-based ensemble collaborative search mechanism.When constructing the ensemble model,the ensemble combination and the decision tree structure are dynamically optimized according to the combined effect between the decision trees based on the evolution of the underlying decision trees.The synergistic mechanism between the two evolutionary searches improves the classification capability and interpretability of the ensemble while reducing the complexity of the ensemble system.The performance of the proposed method is validated with 22 classification datasets.The results show that the dual evolutionary mechanism can effectively reduce the ensemble redundancy structure,improve the ensemble generalization performance,and construct decision trees with lower complexity and better combination effect.(3)To address the current diagnostic problems in clinical medicine,researchers have applied the dual evolutionary ensemble learning algorithm to a study of early diagnosis and postoperative recurrence prediction of breast cancer in Wisconsin.The experimental results show that the ensemble model constructed by the dual evolutionary mechanism has a good classification effect and can effectively improve the interpretability of the model by removing redundant structures.The decision rules extracted from the model can be adjusted and updated according to the actual situation,providing a solution with accuracy and interpretability for the physician’s diagnostic process.This paper completes the study of the dual evolutionary ensemble learning method for classification problems and has some practical value in medical issues through the above research.In particular,the dual evolutionary ensemble method does not require pre-setting the ensemble structure and realizes the automatic construction of the ensemble,thus reducing the interference of human decision-making.