Machine Learning Algorithms for Diabetes Diagnosis Prediction

Diabetes is one of the most common health conditions in the United States, affecting more than 37 million U.S. adults. Despite how many people are affected by this disease, a permanent cure does not currently exist. Therefore, it is very important for patients to get the proper diagnosis and treatment. If diabetes is not diagnosed and properly treated to control blood sugar levels, serious health problems can occur. Though undiagnosed cases are very dangerous, still 1 in 5 U.S. adults with diabetes do not know that they have it. Therefore, a method for early detection or prediction of the disease is imperative for healthcare workers to provide timely distribution of treatment. Presently, hospitals collect large volumes of patient data with various tests and provide appropriate treatment based on diagnoses. With big data analytics, hidden patterns can be found in the data that help to predict diagnosis outcomes. Such predictions can help lower the rate of undiagnosed patients. In our paper, we propose a prediction model for classification of diabetes based on various features of patient data. We developed and applied six machine learning algorithms on two diabetes data sets to predict diagnoses. We found that the best algorithms for prediction with our two data sets were Ridge Classifier with an accuracy of 83.12%, along with Random Forest and Decision Tree classifiers, both achieving accuracies of 95%.