FIGURE 4.

Model construction and data processing. The whole project was composed of model construction, validation, and inflammatory bowel disease unclassified reclassification. The study recruited 239 pediatric patients from the Genetics of Pediatric Inflammatory Bowel Disease study at Southampton Children’s Hospital. The clinical data were used to search the best parameters for classification and train the model. The linear SVM was applied to construct the optimal model, allowing for assessing the relevance with the disease of the selected variable. Then, the optimal penalty parameter (C) tuning and fivefold cross-validation scheme (RFE-CV) help maximizing the classification accuracy by avoiding overfitting. Adapted with permission from Mossotto et al. (2017), Nature Publishing Group.