Abstract
BACKGROUND: Budd-Chiari syndrome (BCS) is a rare global condition with high recurrence rates. Existing prognostic scoring models demonstrate limited predictive efficacy for BCS recurrence. This study aims to develop a novel machine learning model based on multiple kernel learning to improve the prediction of 3-year recurrence in BCS patients. METHODS: Data were collected from BCS patients admitted to the Affiliated Hospital of Xuzhou Medical University between January 2015 and July 2022. The dataset was divided into training, validation, and test sets in a 6:2:2 ratio. Models were constructed by evaluating all combinations of four kernel functions in the training set. Hyperparameters for each model were optimized using the particle swarm optimization (PSO) algorithm on the validation set. The test set was used to compare kernel function combinations, with the area under the curve (AUC), sensitivity, specificity, and accuracy as evaluation metrics. The optimal model, identified through the best-performing kernel combination, was further compared with three classical machine learning models. RESULT: A kernel combination integrating all four basic kernels achieved the highest average AUC (0.831), specificity (0.772), and accuracy (0.780), along with marginally lower but more stable sensitivity (0.795) compared to other combinations. When benchmarked against classical machine learning models, our proposed MKSVRB (Multi-Kernel Support Vector Machine Model for Three-Year Recurrence Prediction of Budd-Chiari Syndrome) demonstrated superior performance. Additionally, it outperformed prior studies addressing similar objectives. CONCLUSION: This study identifies risk factors influencing BCS recurrence and validates the MKSVRB model as a significant advancement over existing prediction methods. The model exhibits strong potential for early detection, risk stratification, and recurrence prevention in BCS patients.