Fig. 4.

Fourteen features important in LASSO logistic regression model to distinguish IM lipomas from WDLSs: A selecting an optimal value of tuning parameter ($\lambda$) in the LASSO logistic regression model was conducted using tenfold cross-validation. The misclassification error was plotted against $log(\lambda )$. $\lambda$ of 0.019 ($log(\lambda )$ = − 3.96) was selected according to tenfold cross-validation. The green dash vertical line denotes the optimal value using minimum criteria; B fourteen features’ importance was obtained using the LASSO logistic regression model. The bar chart of the absolute standardized coefficients showed the feature importance ranking; and C receiver operating characteristic (ROC) curves were plotted for learning and testing data sets showing the area under the curves (AUCs) obtained using the LASSO logistic regression model