Directed cyclic graph-based feature selection and modeling of the dampness syndrome of chronic gastritis

This study aimed to investigate the feasibility of the directed cyclic graph (DCG) in the feature selection and modeling of dampness syndrome to objectively diagnose chronic gastritis (CG). The diagnostic information of patients with dampness syndrome was selected from 919 cases collected in our previous study. Relevant characteristic variables were chosen using the combined rough set and mutual information (RS-MI) method. These selected variables were then used to construct a DCG model. The selected variables were consistent with the symptoms described in traditional Chinese medicine (TCM). The classification accuracies of both dampness syndromes were determined through DCG modeling. The accuracies of the dampness-heat accumulating in the spleen-stomach and the dampness obstructing the spleen-stomach were 90.4% and 78.7%, respectively. Therefore, the DCG model was superior to Navie Bayes(NB) model in terms of classification ability. The classification accuracy rate of the DCG model of the dampness obstructing the spleen-stomach was higher by 1.1% than that of the NB model. In conclusion,feature selection and model construction methods can be used to objectively evaluate the TCM syndromes of CG; nevertheless, these methods should be further investigated and promoted.