Predicting Diabetes Mellitus With Machine Learning Techniques

Diabetes mellitus is a chronic disease characterized by hyperglycemia.Patients may suffer from multiple complications if they are not treated promptly.At present,the prevention and treatment of diabetes in China has shown a ‘three lows’ trend,that is,low informed rate,low treatment rate,and low cure rate.In-depth research on diabetes and related markers can help increase informed rate,treatment rate,as well as the development of drugs.With the continuous improvement of machine learning,smart medical care has become a hot research topic.This study focuses on physical examination data and diabetes protein markers.In Chapter 3,this study use machine-learning methods to learn the two sets of physical examination data,and select important features to build an efficient diabetes diagnosis model.In Chapter 4,this research uses ensemble methods to study diabetes protein markers,including combined feature extraction methods and ensemble classifiers.The main work and innovations of this study are as follows.(1)Establish diabetes prediction models based on physical examination data.This chapter mainly studies the Luzhou medical examination data set and the Pima Indians data set,uses random forest,neural network,and decision tree to build prediction models,and selects features based on principal component analysis and maximum correlation minimum redundancy method.In Chapter 3,modeling based on important diagnostic indicators and random forest method,an efficient diabetes prediction model can be established,which has higher accuracy and shorter running time,and has higher promotion value.(2)Establish a recognition model for diabetes protein markers.The Chapter Four focuses on the study of proteins.This chapter uses ensemble methods to research the problem of protein classification,and finally achieves good results.A comprehensive study of diabetes protein markers is conducive to an in-depth understanding of diabetes and to the development of therapeutic drugs.In chapter 4,three amino acid-based feature extraction methods and six traditional classifiers are used.After integrated processing,four combined feature extraction methods and two ensemble classifiers can be obtained.According to the experimental results,the performance of the ensemble methods are better than that of the single method.