Optimizing Diabetes Prediction with Multi-Layer Perceptron: A Comprehensive Analysis on the Pima Indians Dataset

Diabetes is a health complication that becomes severe or even fatal and is a common condition in the global population, and this makes early diagnosis important. This research examines the ability of a machine learning approach utilizing the MLP classifier to identify the onset of diabetes from the Pima Indians Diabetes dataset. Eight clinical and demographical variables, glucose, blood pressure, body mass index, age, gender, etc., are included in the dataset, which has reasonable clinical and statistical value for building a useful model. All these skewness issues were corrected using a pipeline that involves cleaning data by removing outliers, handling missing values, and normalizing the features. Filter, wrapper, and embedded method were used to estimate the importance of a feature, and hence, less important features like ‘Skin Thickness’ were dropped. In the proposed MLP classifier architecture, only one hidden layer is implemented, including 12 neurons, and using the ReLU activation function by adjusting the hyperparameters, an excellent solution is obtained. The performance of the proposed model was evaluated on the test set with 86.36 % classification accuracy, 86.24 % precision for nondiabetic subjects, 94 % recall for both, 86.67% precision for both, and 72.22% recall for diabetic subjects. These results prove the usefulness of MLP classifiers that can work with imbalanced data and also still possess sufficient predictive power. In fact, this study presents an excellent architecture for feature selection and data preparation to enhance a machine learning model's performance. This has been empowered by the results to inform the deployment of real neural network-based systems for diabetes prediction whilst focusing on scale and interoperability into decision architectures in the healthcare domain. The future work will involve revenue graceful neural structures and realtime implementations of the foregoing diabetes detection model.