Research On Key Technologies For Screening Patients With Hypertension And OSAHS

Obstructive sleep apnea hypopnea syndrome(OSAHS)is a sleep-induced respiratory disease caused by lesions in the upper respiratory tract.The function of each system of the patient may cause different degrees of damage,which may easily induce systemic diseases.Studies have shown that hypertension is closely related to the onset of OSAHS,and the incidence of OSAHS is significantly elevated in patients with hypertension.In the diagnosis and treatment of patients with hypertension and OSAHS,the patient’s OSAHS disease is easily missed and the patient is delayed.The diagnosis and treatment of OSAHS diseases is time-consuming and costly.In recent years,although many researchers have used statistical methods to construct OSAHS patient screening models,researchers have proposed that existing patient screening models are accurate,specific,and sensitive.Low problem.In order to improve the accuracy of OSAHS screening model in hypertensive patients,reduce the missed diagnosis of OSAHS disease in hypertensive patients,and to find the pathogenesis of hypertension-induced OSAHS disease,this study used the machine learning method to collect the first in Yunnan Province.A data set of 398 hypertensive patients in the People’s Hospital Respiratory Sleep Center was used to construct an OSAHS screening model in patients with hypertension.Because the collected data sets have imbalances between classes,feature redundancy,and irrelevance,the data set needs to be pre-processed before the screening model is constructed.Aiming at the imbalance of the sample size between the collected data sets of hypertensive patients,this paper proposes an improved Borderline-Smote unbalanced data processing algorithm.Using the improved Borderline-Smote algorithm,in the process of class-class sample balance processing of data sets,through the fine-grained division of the minority samples in the data set,find the dangerous data samples in a few classes,and then use the dangerous data sample set to generate a new minority.Class samples to achieve sample balance between dataset classes.The data set of hypertensive patients was 398 cases,including 364 patients with hypertension alone,34 patients with hypertension and OSAHS,and the balance was 697 cases,including 364 patients with hypertension alone,hypertension combined with OSAHS.A data set of 333 patients with hypertension.In order to verify the correctness and effectiveness of the improved Borderline-Smote algorithm,the performance of the algorithm is compared with the Borderline-smote1 algorithm,the Borderline-smote2 algorithm and the GSVM traditional unbalanced data processing method.The results of the Accuracy,Precious,Recall and F1 values show that the improved Borderline-Smote algorithm performs significantly better than the original Borderline-smote1,Borderline-smote2 and GSVM unbalanced data processing methods when dealing with unbalanced data sets.Secondly,for the problem of characteristic redundancy and irrelevance in the dataset of hypertensive patients,this study proposes MRMR-SVM-RFE feature selection algorithm.When performing feature selection,the algorithm combines the minimum redundancy maximum correlation feature selection algorithm(MRMR)with the SVM-RFE algorithm to effectively eliminate redundant and uncorrelated features in the data set.Third: By analyzing the characteristics of the dataset of this study,the data pre-processed dataset uses the cascading forest in the deep forest model to construct a screening model for hypertension and OSAHS patients,and in SVM,naive Bayes,decision tree,and perception respectively.The performance of the model was compared using the Accuracy,Precious,Recall,and F1 values on the neural network model.The results showed that the model of hypertension and OSAHS screening model constructed in this study was correct and effective;and the hypertension constructed in this study.The performance of the combined OSAHS patient screening model was significantly better than the screening model constructed using other classification algorithms.