Optimization of an Analysis Method for Diabetes Prediction Using Classical and Ensemble Machine Learning Techniques

Nowadays, diabetes has become a prevalent and significant illness worldwide, causing harm to the circulatory system and leading to complications such as vision loss, kidney problems, and heart disorders. Detecting diabetes early on is crucial in order to implement more effective treatments, control blood sugar levels, and reduce the risk of associated complications affecting both small and large blood vessels. It also provides an opportunity to make lifestyle changes and use targeted medications before irreversible damage occurs in organs and tissues. To achieve this, a method based on CRISP-DM is proposed, which utilizes five traditional machine learning algorithms and ensemble techniques, including RandomForest, DecisionTree, XGboost, Logistic Regression, and Neural Networks. These algorithms are applied to a dataset containing 15,000 records from the National Institute of Diabetes and Digestive and Kidney Diseases [1]. To assess the effectiveness of the predictive models, quality measures such as Accuracy, Precision, Recall, and F1-Score are used for comparison.