Development of a machine learning-based model for predicting risk of early postoperative recurrence of hepatocellular carcinoma

Figure 4 Importance of feature variables analyzed by LASSO regression, XGBoost, and random forest, and top 10 variables ranked in their importance from the highest to lowest and selected. The Venn diagram was drawn by selecting the features common to all three models (taking the intersection). A: LASSO regression with a vertical line was drawn at the value selected using the ten-fold cross-validation, and a minimum mean square error of λ of 0.024 was chosen to obtain the characteristics of the 11 non-zero coefficients; B: The importance of variable features was analyzed using the XGBoost algorithm; C: The importance of variable features was analyzed using the random forest algorithm; D: Venn diagram with eight key characteristic variables at the intersection: Age, intratumoral arteries, alpha-fetoprotein, blood glucose, number of tumors, γ-glutamyl transferase to lymphocyte ratio, liver cirrhosis, and platelets. AFP: alpha-fetoprotein; GLR: γ-glutamyl transferase to lymphocyte ratio; SII: Systemic immune-inflammation index.