Cardiac Diagnostic Feature and Demographic Identification (CDF-DI): An IoT Enabled Healthcare Framework Using Machine Learning

Abstract

The incidence of cardiovascular diseases and cardiovascular burden (the number of deaths) are continuously rising worldwide. Heart disease leads to heart failure (HF) in affected patients. Therefore any additional aid to current medical support systems is crucial for the clinician to forecast the survival status for these patients. The collaborative use of machine learning and IoT devices has become very important in today’s intelligent healthcare systems. This paper presents a Public Key Infrastructure (PKI) secured IoT enabled framework entitled Cardiac Diagnostic Feature and Demographic Identification (CDF-DI) systems with significant Models that recognize several Cardiac disease features related to HF. To achieve this goal, we used statistical and machine learning techniques to analyze the Cardiac secondary dataset. The Elevated Serum Creatinine (SC) levels and Serum Sodium (SS) could cause renal problems and are well established in HF patients. The Mann Whitney U test found that SC and SS levels affected the survival status of patients (p p > 0.05). The Cox regression model also found a significant association of age group with the survival status using follow-up months. Furthermore, the present study also proposed important features of Cardiac disease that identified the patient’s survival status, age group, and gender. The most prominent algorithm was the Random Forest (RF) suggesting five key features to determine the survival status of the patient with an accuracy of 96%: Follow-up months, SC, Ejection Fraction (EF), Creatinine Phosphokinase (CPK), and platelets. Additionally, the RF selected five prominent features (smoking habits, CPK, platelets, follow-up month, and SC) in recognition of gender with an accuracy of 94%. Moreover, the five vital features such as CPK, SC, follow-up month, platelets, and EF were found to be significant predictors for the patient’s age group with an accuracy of 96%. The Kaplan Meier plot revealed that mortality was high in the extremely old age group (χ2 (1) = 8.565). The recommended features have possible effects on clinical practice and would be supportive aid to the existing medical support system to identify the possibility of the survival status of the heart patient. The doctor should primarily concentrate on the follow-up month, SC, EF, CPK, and platelet count for the patient’s survival in the situation.

Keywords

• cardiac disease
• feature selection
• multicollinearity
• machine learning
• IoT