On The Survival Assessment of Diabetic Patients Using Machine Learning Techniques

The extraordinary improvement in biotech and medical sciences have given rise to an impactful data production from stour Electronic Health Records (EHRs), and it has contributed significantly to the Kaggle source from which the data for this research was obtained. The dataset consists of 1416 recorded cases of diabetic patients from 130 various hospitals in the United States. This study thus assesses the survival rate of diabetic patients using machine learning techniques, and determined the duration it will take a diabetic patient to survive based on the application of the most appropriate algorithm. The research tested the application of four different algorithms which include support vector machine, logistic regression, decision tree and k-nearest neighbors’ algorithm. In line with their accuracy measured by f1-score, precision, recall and support metrics; k-nearest neighbors is seen to outperform all other algorithms for predicting the survival rate of the patients. The research also revealed that it takes a diabetic patient 30 days to survive if the patient is placed on medications according to the available information, and that the medication given to the diabetic patients is less effective in the aged patients and more effective among the younger patients.