ASSOCIATION RULE BASED ANOMALY DETECTION IN DIABETIC PATIENTS’ DATA.

Abstract

Discovering interesting patterns within symptoms of diabetes mellitus data remains a major data mining application. This paper presents an association rule based approach towards detecting anomalies in diabetes mellitus patients’ data. This approach extracts interesting frequent symptoms pattern, mines association rules and detects anomalies in the dataset using the mined rules. Diabetes Mellitus is an accumulation of metabolic infections in which a human being has elevated blood sugar due to a number of associated symptoms. These symptoms include regular urination, excessive eating, weight loss, increased need for liquids amongst others.Dataset of diabetes mellitus patients containing 150 records and 48 symptoms is sourced from Obafemi Awolowo University Teaching Hospital, Ile-Ife Nigeria. The method is implemented in Python Integrated development environment. The performance is evaluated based on number of frequent symptoms, mined rules and anomalies detected. The strongest rule recorded from the rule mining has a confidence threshold of 96%. This means that the occurrence of Polydipsia will result to a 96% probability of a presence of Polyuria in a diabetes patient. The result from the anomaly detection shows that an average percentage of 27% anomalies are detected in the diabetes data. The paper shows that anomalies in diabetes mellitus diagnosis can be detected using our approach.