Personalized Diabetes Analysis Using Correlation-Based Incremental Clustering Algorithm

Abstract

Digested food is used as energy; the body breaks down carbohydrates from the food into glucose which is absorbed into the body using hormone insulin. This insulin is helpful in reducing the risk of diabetes. Diabetes Mellitus is a disorder of metabolism; it is one of the highest occurring diseases in the world, having affected over 422 million people. Diabetic level in person depends on various factors; if their values are kept in control, a diabetic patient can improve his/her expectancy of life. The trends in the diabetic levels of a person can be extracted using data mining techniques from individual person’s medical reports. Personalization of these medical reports of an individual will produce apt analysis for diabetologist, considering medical history from inception, showing progress of treatment. Such customized approach is categorically beneficial for not only medical professionals’ like the doctors, nurses, medical practitioners, etc. but also for an individual, their families, researchers and entire society at large. This chapter describes the details about incremental clustering approach, Correlation-Based Incremental Clustering Algorithm (CBICA) to create clusters by applying CBICA to the data of a diabetic patients and observing any relationship which indicates the reason behind the increase of the diabetic level over a specific period of time including frequent visits to healthcare facility. These obtained results from CBICA are compared with results obtained from another incremental clustering approaches, Closeness Factor Based Algorithm (CFBA), which is a probability-based incremental clustering algorithm. ‘Cluster-first approach’ is the distinctive concept implemented in both CFBA and CBICA algorithms. Both these algorithms are ‘parameter-free’, meaning only end-user requires to give input dataset to these algorithms, clustering is automatically performed using no addition dependencies from user including distance measures, assumption of centroids, no. of clusters to form, etc. This research introduces a new definition of outliers, ranking of clusters and ranking of principal components. Scalability: Such personalization approach can be further extended to cater the needs of Gestational, Juvenile, Type-1 and Type-2 diabetic prevention in society. Such research can be further made distributed in nature so as to consider diabetic patient’s data from all across the world and wider analysis. Such analysis may vary or can be clustered based on seasonality, food intake, personal exercise regime, heredity and other related factors. Without such integrated tool, the diabetologist in hurry, while prescribing new details, may consider only latest report, without empirical details of an individual. Such situation is very common in these stressful and time-constraint life, which may affect the accurate predictive analysis required for the patient.