Diabetes Prediction: Insights from the Pima Indians Diabetes Dataset

Diabetes mellitus (Type 2 Diabetes) is a major chronic disease that is a global health challenge, hence accurate early detection methods are a necessity. This study proposes an approach using a Feed Forward Network to predict diabetes cases using the Pima Indians Diabetes Dataset (PIDD). An optimised FNN model with ensemble methods is optimised achieving 75.32% classification accuracy, 78.79% specificity, 69.09% sensitivity, an F1-score of 66.67% and AUC-ROC of 0.80 - this does not outperform some of the learning models that are reviewed in the literature reviewed. Comparative analysis with existing methods has been thoroughly addressed, while also addressing key challenges faced when dealing with medical datasets such as model interpretability and data imbalance. Additionally, the optimised Feedforward Neural Network achieves a breakthrough performance of 99.4% accuracy and an AUC-ROC of 1.00 on the NHANES dataset demonstrating an unprecedented performance with a 26.5% average improvement over PIDD benchmarks. Moreover, the implementation adhered in the study does a good job at demonstrating the effectiveness of FNNs when combined with ensemble techniques for diabetes risk assessment, ultimately contributing to the growing body of research in healthcare using artificial intelligence and underlining the practical and ethical considerations for medical deployment. Thus, the study demonstrates the effectiveness of optimized neural networks in early diabetes detection.