A Comprehensive Machine Learning Framework for Diabetes Prediction: Incorporating Unique Performance Metrics and K-Fold Cross-Validation

Diabetes prediction is critical for early intervention, which helps to avoid significant complications like cardiovascular disease, kidney failure, and neuropathy. This study investigates the efficacy of multiple machine learning algorithms for diabetes prediction utilizing a wide range of evaluation measures such as accuracy, precision, recall, F1-score, Area Under the Curve (AUC), Matthews Correlation Coefficient (MCC), and Cohen's Kappa. To ensure robustness and avoid overfitting, the models were validated with the k-fold cross-validation approach, which divides the dataset into numerous folds for training and testing, ensuring consistent and generalizable findings. The primary objective was to evaluate these models' capacity to effectively categorize diabetes patients, reduce misclassification, and strike a compromise between accuracy (avoiding false positives) and recall (detecting genuine positives). Gradient Boosting regularly outperformed other models in all metrics. Its high accuracy, precision, and memory revealed its ability to correctly identify diabetes and non-diabetic cases while minimizing categorization errors. The outstanding F1-score demonstrated the algorithm's ability to balance memory and precision, while the confusion matrix revealed high true positive (TP) and true negative (TN) rates, as well as low false positives (FP) and false negatives (FN). Furthermore, k-fold cross-validation confirmed the consistency of Gradient Boosting's performance across various data splits, bolstering its trustworthiness. In comparison to other algorithms, Gradient Boosting had the most consistent performance, making it an effective tool for improving diabetes risk prediction. This study demonstrates the potential of Gradient Boosting and k-fold cross-validation in generating more reliable, accurate models for early diabetes diagnosis, allowing doctors to make more informed decisions and improve patient outcomes.