Diabetes is known to be one of the most common diseases in the world today. A significant alternative to traditional diagnostic methods that can be time and resource-consuming is an ML-based approach. With the use of historical health lifestyle and diagnostic data, artificial intelligence can predict someone’s risk of developing diabetes. This study aims to explore the potential of ML models for the prediction of diabetes based on different health and lifestyle factors. Furthermore, using a feature importance analysis, the objective is to determine the factors that affect the most diabetes condition. By evaluating multiple methods, configurations and datasets, current study aims to identify the most effective approach for predicting diabetes risk. For training and prediction, the following 4 models were utilized: Random Forest Classifier (RFC), LightGBM, Decision Tree (DTC) and Naive Bayes. The performance of the models was evaluated on the basis of accuracy, precision, recall and F1-score. The study involved testing these models on two datasets: a smaller one that contains medical attributes (PIMA dataset) and another much larger that incorporates mostly lifestyle factors (derived from the Behavioral Risk Factor Surveillance System Survey from 2022). The results show that the LightGBM and Random Forest models outperform the other evaluated classifiers. LightGBM achieved the highest accuracy of 86% in the larger dataset, highlighting the effectiveness of incorporating lifestyle factors. Furthermore, feature importance analysis revealed key predictors, such as weight, height and mobility difficulties to be those that impact diabetes diagnostic the most.