Abstract
Background
Budd-Chiari syndrome (BCS) is a rare condition worldwide with a high recurrence rate. The existing prognostic scoring models have shown limited predictive efficacy for recurrence of BCS patients.The study aim to establish a more effective machine learning model based on multiple kernel learning for predicting the recurrence of Budd-Chiari syndrome patients within three years.
Methods
The dataset was obtained from patients diagnosed with BCS admitted to the Affiliated Hospital of Xuzhou Medical University between January 2015 and July 2022.The data were divided into training, validation, and test sets in a 6:2:2 ratio. We established respective model based on traversal of all combinations of four kernel functions in training set, and selected best hyperparameters for each model by particle swarm optimization (PSO) algorithm in validation set. Test set was conducted for comparasion of kernel function combinations, with AUC (area under the curve), sensitivity, specificity, and accuracy used as evaluation indexs. The optimal model, utilizing the best-selected kernel combination, was then compared with three other machine learning models to further assess its performance.
Result
A kernel combination incorporating all four basic kernels achieved the highest average AUC, specificity, and accuracy, as well as a slightly lower mean but more stable sensitivity across all combinations. In comparison with other classical machine learning models, our model also achieved significant advantages in performance. Furthermore, it outperformed previous studies with similar objectives.
Conclusion
We have explored risk factors influencing relapse of BCS patients and demonstrated our proposed MKSVRB model is superior to previous prediction methods and other machine learning models, showcasing its significant potential in early detection, determination, and prevention of relapse in patients with Budd-Chiari syndrome.