Predicting the diabetes risk by analyzing symptoms using data mining techniques

Diabetes is a chronic disease with a high rate of prevalence among societies, it causes detrimental effects on individuals, societies, and governments. There is an urgent need to predict the risk of developing the disease before occurring. This study explores the machine learning models' efficiency of predicting this risk.

Machine learning algorithms were used including decision tree, Random Forest (RF), Naïve Bayes, and logistic regression. For performance measures, accuracy, precision, f-measure, and recall were used. Those algorithms were applied on diabetes-related symptoms dataset of 520 instances, and 16 attributes, where the attributes focused on the risk symptoms of diabetes, with the final class labeling whether the patient has diabetes or not.

The use of four different algorithms, with 10-fold cross-validation, and a split of 80:20 ratio, has shown that RF has outperformed other algorithms. As it achieved the highest accuracy in cross-validation of 97.5%, and 95.2% using ratio splitting. Also, the Random Forest has achieved the highest precision, recall, and F-measure of 0.975.

RF is suggested to be the best-intended model to predict diabetes risk along with 10-fold cross-validation. It must be a part of the decision-making along with the provided implications and contributions.