Enhancing Predictive Accuracy of LSTM Neural Networks for Diabetes Risk through K-Fold Cross-Validation: Comparison with K-Nearest Neighbors and Expert Systems

Diabetes is a metabolic disease with a rapidly increasing prevalence worldwide. Developing accurate diabetes risk prediction methods is crucial to enable earlier intervention and optimize healthcare cost management. This study aims to enhance predictive accuracy using the Long Short-Term Memory (LSTM) algorithm in conjunction with K-Fold Cross-Validation. The dataset, sourced from Kaggle, includes 14,257 samples and various health factors. Findings indicate that the LSTM model, with K-Fold Cross-Validation, achieves 80% accuracy in predicting diabetes risk. This model outperforms the K-Nearest Neighbors (KNN) approach, demonstrating LSTM's superior ability to capture complex health data patterns. These results highlight the clinical potential of LSTM models in supporting rapid decision-making for healthcare professionals in diabetes risk detection.