Research And Implementation Of Neuropsychological Test Analysis And HIV-associated Dementia Degree Analysis Method Based On Machine Learning

Acquired Immunodeficiency Syndrome is a malignant infectious disease with a very high fatality rate caused by HIV,the human immunodeficiency virus.At present,there is no vaccine to prevent it,and there is no effective medicine or method to cure this disease.It is one of the most difficult medical problems at present.In the early stage of HIV infection,the nerve cells of patients will produce long-term chronic damage.Even with timely and effective treatment,more than half of the patients still suffer from sensory,motor dysfunction,and neurocognitive impairment,namely HIV-associated neurocognitive disorders.The evolution of HAND is divided into three stages:neurocognitive impairment(ANI),mild neurocognitive disorder,and HIV-associated dementia.There are no clinical symptoms or very mild symptoms in the ANI stage,but there is the earliest stage of HAND neuropathological changes.In this stage,if we can accurately predict the diagnosis and carry out targeted prevention and intervention,it may delay or reverse the pathological process of hand and prevent the occurrence of had,which has important scientific significance and clinical value for improving the prognosis of HAND patients and improving the quality of life of patients.At present,machine learning is widely used in the medical field and has achieved significant results in disease prediction,auxiliary diagnosis and other aspects.The research object of this article is clinical neuropsychological test scales and other medical data,and then using machine learning algorithms to construct a two-class model to predict whether the subject is in the ANI stage.In the process of research,the key technical problems are:small data set and complex characteristic variables.Aiming at the problem of fewer data sets,this paper proposes two solutions.Firstly,the data set is expanded;secondly,an algorithm based on ensemble learning is proposed.The XGBoost classification model can deal with small sample data better.After the model is optimized by five-fold cross-validation,it has better classification performance and generalization ability.Although the XGBoost classification model can get better performance when dealing with small sample data sets,the quality and quantity of features in the data set also affect the performance of the model.In response to this problem,this paper proposes a joint classification model based on GRA-RF-XGBoost.The model uses three different feature selectors:GRA,RF,and XGBoost to comprehensively rank the feature variables.Then,three optimization methods are proposed,and the feature set is simplified according to the reality.The correlation degree between the simplified eigenvalues and the result variables is higher,and the more accurate classification results are obtained.Finally,compared with different classification models,it is found that the GRA-RF-XGBoost joint classification model proposed in this paper achieves the best performance on the data set in this paper.