Early prediction of heart disease using the most significant features of diabetes by machine learning techniques

Abstract

Medical science is witnessing high levels of specialization with doctors specializing in specific areas, say, heart disease (HD), diabetes, nephrology, and the like. In the process, patients have to make multiple visits for treatment of simultaneous ailments. Studies show that there is overlap in causes of different diseases. One such co-existence is observed in patients with diabetes suffering from HD too. In many cases, one precedes the other. Hence, it is worth diagnosing that a patient having a particular ailment is likely to develop another. Artificial Intelligence and machine learning methods are widely used in healthcare.  There are few references to such work using data mining approaches. HD is a primary cause of death worldwide. Studies show that diabetes patients also have HD. This paper aims to identify the association and common risk factors between diabetes and HD - this finding aid in anticipating the HD of a diabetic patient. The authors use proven data mining approaches - logistics regression, decision tree, and random forest to arrive at the most accurate results. The validation is done using unsupervised method: K-means Clustering. The initial investigation demonstrates that body-mass-index (BMI) and age are among the key risk factors for diabetes; and smoking habit, age, gender-male and diabetes (glucose level) lead to HD. 31% of diabetic patients had HD.