Comparison the Accuracy of Hepatocellular Carcinoma Recurrence Prediction Models: LASSO Regression, Random Forest and Support Vector Machine Models

Abstract

The diagnostic effects of Lasso-Cox Logistic regression, random forest (RF) and support vector machine (SVM) combined with hepatocellular carcinoma were compared and analyzed. First, the data quality is improved through preliminary data preprocessing, including the establishment of binary features, categorical variable conversion, missing value processing and lifetime zero value processing. Then COX regression is used to check whether there is a high-dimensional problem in the data, and the stability of the model is improved by multicollinearity test. Then the Lasso regression model is trained to screen significant feature variables and improve the model performance. In the random forest model, the parameters are adjusted by Out-of-Bag (OBB) error rate analysis, and the optimal model is obtained. In SVM model, the optimal penalty parameter and kernel function parameter are determined by cross-validation of optimization parameters. When evaluating the prediction rate of the model, the accuracy rate is taken as the main reference index, and the Lasso-Cox regression, RF and SVM are analyzed in detail. The results show that the RF model in this study has the highest accuracy. The results suggest that RF model may be more suitable for clinical indicators to predict Hepatocellular carcinoma (HCC) recurrence. In the future, larger data sets can be utilized for validation, and the model content can be improved to further optimize and improve the prediction accuracy. This study provides a reference for the selection of HCC recurrence risk prediction model, and is expected to provide a certain degree of reliable support for clinical judgment.