Diabetes Diagnosis via XCS classifier system

This study aims to use novel concepts of artificial intelligence to design an expert clinical system. This system is able to diagnose the diabetes disease at the right time automatically. Target population in this research was existing information on the web site of California University including 768 patients. Volume of samples was taken 500 in this study. Selection of the patients was conducted randomly and the data required for designing this system were extracted accordingly. The expert system developed in this paper was a learning system as an improved version of eXtended Classifier Systems (XCS). Extended classifier systems are known as one of the most successful learning agents in this field of artificial intelligence. They are comprised of a set of simple rules with “if-then” format. Each rule predicts a particular reaction (i.e. type of disease) regarding the information received from the environment. This set of rules is “evolved” interacting with real data, while their prediction accuracy is gradually enhanced. This evolution is usually done using the patterns inspired from the nature such as genetic algorithm. In this research, the system started to learn by application of a real dataset collected. It performance was then examined on some 268 other patients, the results of which were compared with some conventional data mining methods. This comparison indicates preference of the proposed method with other techniques in terms of prediction accuracy. Installation of these systems in hospitals and application of them as a handy tool for physicians can improve decision-making process for diagnosis and provide more comfort for the patients.