Pattern Classification of Foot Diseases using Decision Tree

The purpose of this study was to investigate significant knowledge from developing the prediction model for pattern classification of foot diseases using decision tree based on clinical data stored in the Foot Clinic. Sample data of 80 patients diagnosed with a single disease in 1267 patients. Dependent variable was composed of 8 diseases. Independent variables were selected with 28 variables in the whole 37 attributes closely related to disease. The whole data was divided into training data and test data, and prediction rate was verified by C5.0, C&RT, QUEST and CHAID algorithm. As the result, the C5.0, the top of prediction rate, is applied for final analysis of disease category. The importance factors related closely with 8 disease were RCSP, right pelvis elevation and pelvis rotation condition. In the case of both pes planus and left pes planus, it were shown same result that left RCSP was higher than right RCSP. However, the left pes planus had possibility to accompany the body unbalance by pelvis rotation and pelvis elevation difference. Pes cavus was always shown if any feet had above 1° for RCSP, and gastro-soleus muscle tightness was appeared if there was pelvis elevation difference with similar RCSP on both foot. In conclusion, we were able to conclude that factor of each node, including RCSP, which formed decision-tree, major diagnosis for distinction lower limbs disease and obtain the 8 diagnosis rules. Key-Words: Biomechanical analysis, Custom-made insole, Pes cavus, Foot pressure, EMG