Diabetes mellitus, a chronic metabolic disorder, affects millions globally and poses a significant public health challenge. Early detection and timely intervention are crucial to prevent complications and improve patient outcomes. The increasing digitization of healthcare and the availability of extensive patient datasets have paved the way for intelligent, automated, and scalable diagnostic systems. In this study, a novel Artificial Intelligence and Machine Learning (AIML)-driven deep learning framework is proposed for enhanced diabetes detection using an optimized Bidirectional Gated Recurrent Unit (BGRU) model. The architecture is specifically tuned using the Self-Adaptive GlowWorm Swarm Optimization (SA-GSO) algorithm to achieve superior accuracy and generalization. The PIMA Indian Diabetes dataset is employed for model training and validation. It contains clinical features of female patients, which are carefully processed and refined through multiple stages. First, data preprocessing is undertaken using multivariate imputation via chained equations (MICE), ensuring data completeness and integrity. Feature scaling standardizes the attributes, allowing the learning algorithms to converge more effectively. To address the challenge of high dimensionality and redundant information in patient data, feature selection is performed using the Student PsychologyBased Optimization (SPBO) algorithm. SPBO evaluates the significance of each feature in relation to the prediction objective, thereby selecting only the most impactful inputs. Subsequently, the optimized BGRU model is deployed for classification. BGRU, being a variant of recurrent neural networks, captures sequential and temporal dependencies in the data while mitigating the vanishing gradient problem inherent in traditional RNNs. The bidirectional nature of BGRU enhances its ability to consider both past and future contextual information, making it wellsuited for structured medical data. However, the architecture’s performance heavily relies on hyperparameter settings, such as learning rate, dropout rates, number of neurons, and batch size. To optimize these parameters effectively, the SA-GSO algorithm is employed. Inspired by the intelligent foraging behavior of glow-worms, this self-adaptive metaheuristic balances exploration and exploitation of the parameter search space. It dynamically adjusts search intensity based on solution quality, promoting faster convergence and better solutions. When integrated with the BGRU framework, SA-GSO significantly enhances classification accuracy and reduces computational overhead. Comparative analysis with standard models such as CNN, AE, DBN, and traditional machine learning classifiers like MLP and XGBoost indicates that the proposed model delivers superior results across all major metrics including accuracy, precision, recall, and F1-score. The research not only contributes a high-performing hybrid deep learning model but also establishes a workflow for efficient clinical data preprocessing, feature reduction, and intelligent hyperparameter tuning. The outcomes suggest a robust, scalable, and interpretable diagnostic aid for healthcare providers. This predictive tool can be easily integgrated into telemedicine applications, remote health monitoring systems, and clinical decision-support platforms. Moreover, the methodology adopted in this work offers a foundation for further research. Future enhancements could explore the inclusion of additional clinical attributes such as genetic markers, lifestyle data, and medication history. Investigating the performance of the model across different population cohorts would help validate its generalizability. Integration with real-time data from wearable sensors and IoT devices can also be pursued to enable proactive and personalized healthcare delivery.

Artificial Intelligence in Medicine, Diabetes Detection, Bidirectional Gated Recurrent Unit (BGRU) , Self-Adaptive Glow-Worm Swarm Optimization (SA-GSO) , Feature Selection & Student Psychology-Based Optimization (SPBO)