Analysis of pediatric foot disorders using decision tree and neural networks

Data mining is method to extract hidden predictive information, and it has been recognized by many studies. The object in the study was to discover meaningful knowledge between the foot disorder and biomechanical parameters related to symptom using C5.0 decision tree and neural networks. The first medical record data of 174 pediatric patients was extracted for analysis, in total 279 records, and they were diagnosed with a complex foot disorder. The dependent variable consists of five complex disorder groups, and 14 independent variables related to disorder groups were selected by importance, in 34 variables. The extracted data was separated to generate an ideal prediction model. After development of the prediction model, the prediction rate was verified and neural networks were applied for analysis of predictor importance and classification prediction. Consequently, a major symptom information in 13 diagnosis patterns was confirmed.