Research And Application Of Feature Selection Method Based On Heuristic Algorithm

In recent years,the incidence of depression is increasing year by year,and the cost of treatment is increasing too.In current life,self-assessment and some indicators of hormone are used to diagnose patients with depression.Self-assessment is subjective and easy to deviate from objective facts;the hormone indicators are usually affected by some physical diseases,it is easy to cause misdiagnosis.At present,with the in-depth study of EEG in academia,more and more scholars combine the knowledge of machine learning and EEG,by the feature extraction,selection and classification of EEG,the recognition model of depression is finally established,it can provide an objective aid for clinical diagnosis.This paper focuses on the feature selection step in the process of machine learning,by combining the traditional feature selection and the heuristic algorithm,a fusion feature selection algorithm FM_GASA is proposed,we apply it to the research of depression recognition based on EEG in order to improve the reliability of recognition.The contents and results of this paper are as follows:1.We propose the fusion feature selection algorithm FM_GASA,this algorithm takes into account the advantages and disadvantages of both Filter and Wrapper feature selection.Firstly,the two Filters named Variance Analysis and Mutual Information are used to search in the feature space,then the Wrapper algorithm performs a secondary search on the filtered space.In the search strategy of Wrapper algorithm,for the premature convergence of Genetic algorithm,local optimization is performed for each individual in the population,and the simulated annealing strategy is introduced in the process of optimization,finally an improved Genetic algorithm is formed.The whole algorithm is verified on five public datasets in UCI,and three kinds of Filter algorithms and two kinds of Wrapper algorithms are added for comparison,the experimental results show that the fusion feature selection algorithm FM_GASA has a better effect.2.The fusion feature selection algorithm FM_GASA was applied to the research of depression recognition based on EEG,by selecting features of resting EEG and 5-band audio stimulating EEG,a more effective feature combination was created.The experimental results show that the Random Forest Algorithm can achieve 80.4% accuracy in the test set under negative audio stimulation,which is better than the existing literature we mentioned,this also verifies the effectiveness and practicality of the feature selection algorithm FM_GASA,it can improve the accuracy of recognition to a certain extent.