Likelihood Prediction of Diabetes at Early Stage Using Data Mining Techniques

Abstract

Diabetes is one of the fastest growing chronic life threatening diseases that have already affected 422 million people worldwide according to the report of World Health Organization (WHO), in 2018. Due to the presence of a relatively long asymptomatic phase, early detection of diabetes is always desired for a clinically meaningful outcome. Around 50% of all people suffering from diabetes are undiagnosed because of its long-term asymptomatic phase. The early diagnosis of diabetes is only possible by proper assessment of both common and less common sign symptoms, which could be found in different phases from disease initiation up to diagnosis. Data mining classification techniques have been well accepted by researchers for risk prediction model of the disease. To predict the likelihood of having diabetes requires a dataset, which contains the data of newly diabetic or would be diabetic patient. In this work, we have used such a dataset of 520 instances, which has been collected using direct questionnaires from the patients of Sylhet Diabetes Hospital in Sylhet, Bangladesh. We have analyzed the dataset with Naive Bayes Algorithm, Logistic Regression Algorithm, and Random Forest Algorithm and after applying tenfold Cross- Validation and Percentage Split evaluation techniques, Random forest has been found having best accuracy on this dataset. Finally, a commonly accessible, user-friendly tool for the end user to check the risk of having diabetes from assessing the symptoms and useful tips to control over the risk factors has been proposed.