Research On Weights Optimization In Classifiers Based On Hybrid Rice Optimization Algorithm

Classification is one of the most important tasks in machine learning,while the performance of classifier is mainly influenced by structure and parameter.Among them,the optimization of parameter is still a difficult problem to solve.In order to improve the performance of classification algorithms,many parameter weight optimization algorithms have been proposed.Traditional optimization algorithms,such as gradient descent,are still prone to fall into local optimum and have poor convergence.Usually the optimization of weight parameters requires searching for the optimal solution in the real range,which is a very high computational complexity problem.The intelligent optimization algorithms based on probabilistic search mechanism such as differential evolution algorithm and particle swarm optimization algorithm have good optimization performance,which have been successfully applied to solve such problems.However,these problems have not been completely solved.As a newly proposed intelligent optimization algorithm,hybrid rice optimization algorithm based on heterosis theory has fast searching speed and strong searching ability.As a result,this thesis attempts to use the hybrid rice optimization algorithm for weight optimization problems in machine learning classifiers to explore its potential in machine learning applications.The main tasks of this thesis are as follows:1.The feature weights of machine learning classifiers are optimized by hybrid rice optimization algorithm.Attribute weighted K-nearest neighbor classifier based on hybrid rice optimization algorithm and attribute weighted Naive Bayes classifier based on hybrid rice optimization algorithm are proposed.Comparative experiments between the original classifiers and the improved classifiers based on intelligent optimization algorithms are carried out.The results show that the feature weights of K-nearest neighbor classifier and Naive Bayes classifier optimized through the hybridization and self-crossing breeding mechanism of hybrid rice optimization algorithm train better performance classifiers.2.Hybrid rice optimization algorithm is utilized to optimize the weight of the classifier integration,thereby improving the generalization performance of the classifier.After measuring the difference of the base classifiers,the base classifiers with larger difference are selected.Considering the search ability,hybrid rice optimization algorithm can search for the optimal weight vector and obtain a classifier integration model with higher accuracy.3.The proposed methods are applied to solve the image classification in the field of remote sensing,to explore the applicability of machine learning classification algorithms based on hybrid rice optimization algorithm in the field of remote sensing images.In general,this thesis applies hybrid rice optimization algorithm to optimize the weight of K-nearest neighbor classifier,Naive Bayes classifier and classifier integration.Some machine learning public datasets and remote sensing images are used to test the proposed methods.The experimental results show that the proposed methods can effectively improve the classification accuracy and robustness of the classifier,especially when the data sets have many noise samples and redundant features in practical applications.The classification accuracy of the modified classifiers has been improved,which has a certain potential in the performance optimization of machine learning and a wide range of applications in the field of remote sensing images.