Application Of Machine Learning Algorithms For The Diagnosis Of Sarcopenia

In recent years,with the rapid development of smart medical care,the use of artificial intelligence technology to assist medical diagnosis has become a trend,do-mestic and foreign research on machine learning assisted medical care has become more and more mature,and machine learning pre-diagnosis models for many diseases have been established.The use of machine learning algorithms to assist doctors in diagnosis will greatly improve the scientific nature of the diagnosis,effectively over-come the subjective problems caused by doctors' empirical diagnosis,and will reduce the burden on doctors to a certain extent as well.According to the above research background,this article relies on our laboratory research projects to study the dis-ease of sarcopenia and try to establish a machine learning model which suitable for the diagnosis of sarcopenia.Some relevant theoretical basis are detailed introduced in the second and third chapter,mainly including data processing,model evaluation and algorithms introduction.For the selection of machine learning algorithms,this paper will use support vector machine(SVM),random forest(RF),XGBoost,and CatBoost four algorithms to make a comparative study.For the evaluation index of the model,this paper chooses the accuracy rate,precision rate,recall rate and F1 values.It is worth mentioning that the sarcopenia data in this article contains some behavioral data,in addition to some outpatient data.In Chapter 4,we demonstrated the importance of these behavioral features to the establishment of the pre-diagnosis model,mainly from two aspects,one is the correlation study of features,another is the feature importance given by machine learning algorithms.These behavioral data also enriched the patient's data types.This paper also emphasized the importance of category balance for algorithm training in the classification problem,and used SMOTE algorithm to handle with the category imbalance problem.Compared the model performance before and after the balanced processing,we found that in the processed models,every evaluation in-dicators have been significantly improved,and the specific improvement percentages are given in Table 10 in Chapter 5 of the article.Among the four indicators,the CatBoost algorithm has improved by 2.76%,36.36%,82.50%,66.21%respectively.Finally,four algorithms were applied to the sarcopenia dataset.Through the horizontal and vertical comparison of the effects of each algorithm,and by observing the four indicators,accuracy,precision,recall,and F1 value,the CatBoost algorithm was proved to be a better-performing algorithm.take the data characteristics into consideration,we suggested to use CatBoost machine learning algorithms for pre-diagnosis of the disease on the sarcopenia dataset.Last but not least,the article also compared the performance of several algo-rithms on various data sets,and found the CatBoost algorithm was more stable.By observing the running time of different algorithms we gave a correlate suggestion of the algorithm selection when study other problems.