Diabetes Prediction Using Classical Machine Learning and Deep Neural Networks: A Comparative Study

Diabetes mellitus is a significant is a worldwide health issue that necessitates prompt detection to avert serious consequences. This study examines the performance of five classical machine learning models and a novel Deep Neural Network (DNN) for binary diabetes prediction. The dataset comprises clinical and demographic attributes such as age, BMI, hypertension, and HbA1c levels. Categorical variables were transformed and all features standardized prior to model training. Models evaluated encompasses Logistic Regression, Random Forest, Support Vector Machine, K-Nearest Neighbours, and Gradient Boosting. Each model was trained on $80 \%$ of the data and validated on the remaining $20 \%$, using accuracy and classification metrics to gauge performance.The DNN attained the highest accuracy of $97.23 \%$, surpassing all traditional methods. Gradient Boosting and Random Forest trailed closely with accuracies of $97.21 \%$ and $97.01 \%$, respectively. The DNN’s superior performance can be attributed to its capability to confine complex, non-linear patterns and apply regularization techniques. These results highlight the promise of deep learning in advancing disease prediction and facilitating early clinical interventions.