Bio-marker Detection for Type 1 and Type 2 Diabetes using Deep Learning

Abdul Azeez K, Aravindan M, Adhirai Nandhini A, Tejeswinee K

In developing countries like India, non-communicable diseases such as diabetes have already replaced communicable diseases as the major cause of death. According to data from the International Diabetes Federation(IDF) and 14 cohort studies (representing more than 60 percent of the world population with type 2 diabetes), researchers estimated the burden of type 2 diabetes in 221 countries and territories between 2018 and 2030 and IDF pegs the number of patients with diabetes in India at 65.1 million (it was 50.8 million in 2010) and the number is expected to cross 100 million by 2030 .The number of adults with type 2 diabetes is expected to rise over the next 12 years due to ageing, urbanization, and associated changes in diet and physical activity. In this paper the authors focus on diagnosis of diabetes using the various machine learning techniques of data mining. And, authors have compared various classification techniques such as Naive Bayes, KNN, Adaboost, SVM, Decision tree algorithm J48,Random forest. And three well-performing feature selection algorithms namely, Correlation Feature Subset Selection (CFS), Information Gain(IG) and Gain Ratio (GR) are used to obtain the optimal features contributing to the  diabetes disease. Further, Incremental Feature Selection(IFS)  techniques are applied to further reduce the feature subset from the optimal feature set.