The Classifier Ensemble Pruning Algorithm Based On A Geometric Framework

In the era of rapid development of information technology,the scale and complexity of data are showing exponential growth,and traditional computing methods are no longer able to handle such a large amount of data and changing data structures.Using machine learning methods and data mining techniques to automatically discover and extract useful information and knowledge from massive data is a highly challenging task.Ensemble learning is a method in machine learning that can improve model performance and stability,and is widely used in various fields.However,compared to a single classification model,ensemble learning requires more resources and time to train multiple classification models.In order to reduce the time and spatial complexity of ensemble learning,this thesis uses ensemble pruning method to select a smaller number of base classifiers to achieve better performance.On the basis of geometric framework theory,ensemble pruning research was conducted and a geometric framework based ensemble pruning system was designed.The specific research work is as follows:(1)In order to improve the efficiency and generalization performance of ranking-based ensemble pruning algorithms,a geometric framework-based ensemble pruning algorithm is proposed.This algorithm takes into account both the performance and diversity of the base classifiers and uses Euclidean distance to calculate both factors.Compared to similar ensemble pruning algorithms,the proposed algorithm avoids the problem of using hyperparameters to balance the performance and diversity of the base classifiers during the ensemble pruning process.Additionally,it effectively avoids overfitting issues and enhances the generalization ability of the ensemble pruning model.Experimental results demonstrate that this algorithm achieves improvements in both the efficiency and generalization performance of the ensemble pruning model compared to current algorithms in the same category.(2)To enhance the generalization capability and efficiency of the model,a cluster ensemble-based ensemble pruning algorithm is proposed to reduce redundant base classifiers in ensemble learning.Firstly,an evolutionary clustering algorithm is employed on top of ensemble pruning to enhance the robustness and stability of clustering results by integrating multiple clustering outcomes.Additionally,a random function is utilized to increase the diversity of the initial population.Then,individuals in the population are selected and eliminated,ultimately obtaining a globally optimal clustering result with maximum fitness.Finally,based on the obtained optimal clustering result,all base classifiers are grouped,and classifiers are selected from each group according to the geometric framework theory.Experimental results demonstrate that the proposed algorithm exhibits improvements in accuracy and robustness.(3)A geometric framework-based classifier ensemble pruning system is designed and implemented.Using Vue and ElementUI to implement user interaction interfaces,using Java to implement system business logic,and Python to implement multiple ensemble pruning algorithms.The system includes login module,data management module,algorithm running module,and ensemble pruning result display module.The system can effectively reduce the time and space cost of ensemble learning.