Predictive Modeling of Diabetes Using Multimodel Machine learning and Deep learning Approaches

Abstract

This study discusses the implementation and evaluation of various machine learning algorithms along with one deep learning model for predicting diabetes based on patient medical data. The dataset underwent Preprocessing steps including categorical feature Encoding, feature scaling, and train-test split. The algorithms compared in this study include Logistic regression, Decision Tree, Random Forest, and K-Nearest Neighbors (KNN). Additionally, a Multilayer Perceptron (MLP) model was developed using Keras to explore a deep learning approach with the use of epochs and batch size. The model performance was evaluated using accuracy, precision, and recall metrics, along with learning curve visualizations to analyze model convergence during training. The evaluation results showed that the Random Forest model achieved the highest accuracy among traditional algorithms, while the MLP provided competitive results with strengths in generalization. Visualization of loss and accuracy per epoch offered deeper insight into model behavior throughout the training process. This study demonstrates that a combination of proper data Preprocessing techniques and appropriate model selection significantly influences prediction accuracy. The findings may serve as an early reference for the development of data-driven medical prediction systems and support computer-assisted clinical decision-making (clinical decision support systems).