Kernel Regression Based Optimal Deviation Search Method And Its Application In Age Estimation

Age is strongly associated with the development of disease.With age,the gradual loss of physiological integrity leads to impaired function and increased susceptibility to diseases such as cardiovascular disease,diabetes mellitus and Alzheimer's disease.Usually,these diseases have accelerated damage to the tissue structure and physiological function of the corresponding part many years ago before they are diagnosed,resulting in a certain deviation between the physiological age and the real age,that is,age deviation.Since the physiological age can be obtained by estimation,the research of age estimation method based on deviation search has certain theoretical value and practical significance.At present,age estimation has become a hot and difficult research topic at home and abroad.Studies have shown that age information can be extracted through medical data mining.Many scholars have carried out relevant studies and achieved some results in assisting disease detection and diagnosis.However,there are still some problems in the existing age estimation methods: 1)The traditional age estimation methods make the estimated age approximate to the real age by minimizing MAE value.The age estimation error of the normal control is small,but the classification ability of the estimated age is poor.2)There is a close relationship between age and heart disease,but there is little public research on heart age estimation and the classification of heart disease.3)The pathological age estimation method proposed by our research group does not take into account the age estimation error of the normal control,and its correlation with the traditional age estimation method is still unclear.To solve the above problems,the optimal deviation search method for age estimation is firstly studied,aiming at improving the performance of the existing age estimation methods.Secondly,the heart age estimation algorithm based on the optimal deviation search method is studied,which is used to estimate the heart pathological age.Finally,the correlation between the pathological age estimation method and the traditional age estimation method is studied,and a comprehensive pathological age estimation algorithm is proposed.The main contents of this paper are as follows:(1)A wrapper deviation search method based on kernel regression is proposed.The method uses separability distance and correlation coefficient as evaluation criterion to guide the age deviation search,and searches for the optimal age deviation by maximiz-ing the separability distance or correlation coefficient.The classification ability of the estimated age can be effectively improved.(2)A heart pathological age estimation algorithm based on wrapper deviation search and Support Vector Regression(SVR)is proposed.Firstly,the traditional age estimation method is extended to the heart disease dataset to realize the heart physiological age estimation algorithm,and it is used as the control algorithm.Then,the heart pathological age estimation algorithm based on the kernel regression optimal deviation search method is studied.This algorithm introduces the separable distance and correlation coefficient to design the fitness function,and searches for the optimal age deviation by maximizing the fitness function value.Compared with the control algorithm,the classification ability of the estimated age can be significantly improved.(3)A comprehensive pathological age estimation algorithm based on wrapper deviation weighted kernel regression method is proposed.Firstly,the traditional age estimation module and the pathological age estimation module are weighted together,and the optimal weights are obtained by minimizing the MAE value between the weighted age and the real age of the normal control,the importance proportion relationship between the two methods is deeply understood.Then,this algorithm was applied to three common datasets of heart disease,diabetes and Alzheimer's disease.The results show that the proposed algorithm achieves a good balance effect.It not only improves the classification ability of the estimated age,but also reduces the age estimation error of the normal control.In this thesis,a new idea for the study of age estimation is provided and a new theoretical foundation for improving the classification ability of the estimated age is laid...