| With the improvement of people’s living standards,the prevalence of coronary heart disease is getting higher and higher,and the affected group is further younger.Although traditional medical methods can identify most patients with coronary heart disease,they are costly,inefficient,and rely on the rich experience and professional level of doctors,and also need to make judgments based on many complex medical indicators,which increases the burden of medical services.Therefore,using machine learning to assist in diagnosing whether you have coronary heart disease is a simple,convenient and efficient method.Based on this,this thesis uses machine learning algorithms to construct a classification model of coronary heart disease for auxiliary diagnosis of coronary heart disease.Firstly,based on a coronary heart disease dataset on the Kaggle platform,this thesis preprocesses the original data,and visually analyzes the data in order to understand the impact of each characteristic variable on the prevalence of coronary heart disease.Secondly,the ratio of data with and without coronary heart disease in the dataset is about1:11,which is unbalanced.In order to improve the effectiveness and stability of subsequent models,the imbalanced data was processed by ADASYN algorithm,SMOET algorithm and SMOTE-ENN algorithm,and four models of logistic regression,decision tree,KNN and random forest were established.In order to improve the generalization ability and learning ability of the model,the embedded method was used to select the features,and the feature importance ranking was obtained by establishing three models: decision tree,random forest and Ada Boost,and finally all 17 feature variables were determined for subsequent modeling analysis.Then,aiming at the research problem of auxiliary diagnosis of coronary heart disease,three single models of decision tree,logistic regression and KNN,and three ensemble models of Ada Boost,light GBM and random forest are constructed.Comparing the F1 value and AUC value of the test set,it was found that the F1 value and AUC of KNN reached more than 0.9 in a single model,which had the best effect;the F1 value and AUC value of the three ensemble models were higher than those of the single model,among which the F1 value and AUC of random forest reached 0.9627 and 0.9944,and the classification effect was the best.Finally,the Stacking algorithm is used to establish a fusion model for six models in nine combinations.Comparing all model evaluation indicators,it is found that the F1 value and AUC value of the stacking fusion model are higher than the homogeneous ensemble model overall,among which the Stacking fusion model established by random forest,Adaboost and Light GBM as the primary learner and KNN as the secondary learner has the best generalization ability,and the model does not have the problem of overfitting,which has strong robustness. |
