Comparing of feature selection and classification methods on report-based subhealth data

Sub-health is a state between health and disease conditions, which is common among people living with the fierce competition and rapid pace of modern life. At present, there are no unified approaches to diagnose the sub-health patients. Self-reporting, the use of questionnaires, is one of the most popular approaches to evaluate health conditions. While a questionnaire consists of as many as 400 questions, people are likely to lose patience. This paper presents a machine learning method to mine the sub-health related questions and then provide classification suggestion based on the self-reporting data collected from Sub-health Condition Identification and Classification Research project. To study the most effective mining approaches, four different feature selection methods were applied to discovery the internal relationship among questions and four different supervised learning classifiers were utilized to investigate the most related questions to the specific diagnostic tasks. Experimental results show that artificial neural network achieves the best performance and the final diagnostic accuracy reaches 84.07% with 20 most related questions.