Lasso And Its Related Optimization Models In Clinical Prediction

With the deep integration of big data and clinical medicine,statistical models are gradually applied to clinical prediction.In traditional clinical prediction,single factor variance analysis,logic regression,random forest and support vector machine are used.However,because of the large number of clinical data features and multiple collinearity,the prediction effect of the model is restricted due to the difficulty of feature selection.Therefore,a universal model with feature screening function is needed to solve this problem.Lasso and its related optimization model are used in the clinical prediction.The principle of Lasso is deduced and the reason why Lasso has the feature screening function is discussed.Based on the simulation data,the comparison is made between the method and some common feature selection methods.At the same time,the real clinical data is used to train Lasso to get the prediction model,and then established the Lasso related optimization model according to the model fusion idea to further optimize the prediction performance.Through two experiments in this paper,Lasso has universality in feature selection,and has stable effect on different data models;Lasso has excellent prediction results based on early diabetes risk data.The elastic network model and Lasso logistic model optimized for some limitations of Lasso do achieve better prediction results.Therefore,Lasso is feasible in clinical data.