Diabetes is a chronic metabolic disorder that heightens the risk of complications for women and presents diagnostic challenges owing to imbalanced datasets and the need for interpretable predictive models. In this study, we propose a 1D Convolutional Neural Network (1D CNN) model that achieves an accuracy of 98.61% on German Patient Dataset, comprising 2,000 samples, and 99.35% on the Bangladeshi Patient Dataset, which includes 465 samples. Our model effectively addresses class imbalance by integrating the Synthetic Minority Over-sampling Technique and Edited Nearest Neighbor (SMOTE-ENN), which significantly enhances performance. Additionally, we conducted a statistical comparison with Multi-Layer Perceptron (MLP), Long Short-Term Memory (LSTM), and Bidirectional LSTM (BiLSTM) models, demonstrating our CNN’s superior accuracy while maintaining reduced complexity and enhanced transparency through the integration of SHapley Additive exPlanations (SHAP). Our SHAP analysis revealed significant variations in feature importance between the two populations, offering culturally relevant insights into the risk factors for diabetes. The SHAP analysis not only facilitates interpretability by allowing healthcare professionals to understand the influence of individual features but also emphasizes the cultural context of diabetes risk. Overall, our findings surpass existing methodologies in terms of accuracy and complexity while underscoring the critical need for demographic diversity in predictive healthcare models, paving the way for more effective diabetes prediction strategies.

Convolutional Neural Networks (CNNs) are widely utilized for their robust feature extraction capabilities, particularly in medical classification tasks. However, their opaque decision-making process presents challenges in clinical settings, where interpretability and trust are paramount. This study investigates the explainability of a custom CNN model developed for Covid-19 and non-Covid-19 classification using dry cough spectrograms, with a focus on interpreting filter-level representations and decision pathways. To improve model transparency, we apply a suite of explainable artificial intelligence (XAI) techniques, including feature visualizations, SmoothGrad, Grad-CAM, and LIME, which explain the relevance of spectro-temporal features in the classification process. Furthermore, we conduct a comparative analysis with a pre-trained MobileNetV2 model using Guided Grad-CAM and Integrated Gradients. The results indicate that while MobileNetV2 yields some degree of visual attribution, its explanations, particularly for Covid-19 predictions are diffuse and inconsistent, limiting their interpretability. In contrast, the custom CNN model exhibits more coherent and class-specific activation patterns, offering improved localization of diagnostically relevant features. • Enhance healthcare predictions by explaining neural networks using feature analysis. • Identify key spectrogram patterns to enhance transparency in medical diagnosis. • Provide interpretable insights into neural network classification in health-care. • Support clinicians with data-driven explanations for improved medical decisions. • Visualize key features contributing to accurate patient diagnosis and care.