Study On Feature Selection Strategy Based On Improved Dragonfly Algorithm For Biomedical Data

With the continuous advancement of human society,health have gradually become an important demand for human survival and development.Various types of diseases are the primary factors affecting human life and health,so the growth and development of medical care received more and more attention.With the improvement of genome sequencing technology,a large amount of valuable data has been accumulated.The effective mining and analysis for these data can effectively promote the clinical diagnosis.Gene expression data usually has a small amount of data samples,but it contains a lot of redundant information.Faced with such large-scale data,it is difficult for medical personnel to analyze data effectively in a short period of time.Even if the data is modeled by a data analysis algorithm,too much redundant information can easily lead to over-fitting of the algorithm,thus misleading medical personnel to make a wrong diagnosis.As an effective method of data preprocessing,feature selection method is widely used in the processing of medical data.As an efficient data dimensionality reduction technology,feature selection has gained widespread attention in the medical field and has been successfully applied to the diagnosis of disease data.Among them,the Wrapper-based feature selection method has become a hot topic of research because of its higher classification accuracy and flexibility.The performance of this method is mainly affected by the search algorithm,and the impact of different search algorithms on the results is also different.As an emerging search algorithm,dragonfly algorithm has been widely used in parameter optimization,global optimization and other issues.In this paper,the application of dragonfly algorithm in feature selection is improved and applied to the prediction and diagnosis of disease data such as Parkinson,lung cancer and leukemia.The main research results are as follows:(1)Put forward a kind of Parkinson's disease feature selection strategy based on improved dragonfly algorithm.Aiming at the data of Parkinson's disease that has accumulated so far,this paper proposes a feature selection strategy based on the Improved Boundary Dragonfly Algorithm(IBDA).This method is used to analyze and screen the characteristics of Parkinson's data,which is helpful for the rapid diagnosis and treatment of potential Parkinson's patients.Firstly,the initial population generated by Logical regression sequence is updated by DE algorithm.The best individual is selected as the initial population of BDA update,which is further solved by the parallelism,aggregation,separation,foraging and avoiding enemy behaviors of BDA algorithm.In addition,an elite strategy is adopted in the iterative process to maintain the superiority of the population,and the Levy flight is used to avoid the solution to the local optimum.The experimental data comes from the Parkinson dataset published on UCI,and compares the performance of other feature selection algorithms such as IBDA and WOA,BALO,GA,etc.from several indicators.The experimental results show that the proposed IBDA algorithm has more choices in Parkinson's disease characteristics.Good performance.(2)Proposed a feature selection strategy for high dimensional disease data based on optimal dragonfly algorithm.The redundancy problem of high-dimensional disease data has always been one of the important problems to be solved in medical diagnosis work,but it can be effectively dealt with through feature selection technology.This article focuses on the selection of high-risk diseases such as leukemia and lung cancer,and selects effective disease characteristics to help medical personnel perform early diagnosis.This method mainly adds the elite strategy to ensure the optimal solution transfer based on the algorithm,and adds the simulated annealing algorithm,which effectively improves the convergence speed and prediction accuracy of the algorithm and enhances the search performance of the algorithm.In this experiment,the method is compared with some classical methods,such as information gain,chi-square detection,Relief,GA algorithm and MPSO algorithm.The experimental results show that the method has certain performance in high-dimensional data processing.