Healthcare providers deal with a problem in effectively treating and managing diabetes, which requires early and accurate diabetes prediction. To predict the risk of early-stage diabetes, a custom deep learning model combined with Explainable Artificial Intelligence (XAI) techniques, specifically, Shapley Additive Explanations (SHAP) is proposed in this study. Data gathering from a public available data source is included in the method and cleaning to ready them for use in the model. Thus, the objective is to enhance the evaluation criteria by constructing hybrid deep and lightweight models. With a training accuracy of 99.76% and a test accuracy of 98.71%, the proposed deep learning model performs exceptionally well and has good generalization. Furthermore, the model performs better in precision, recall, and F1-score than other machine-learning models. The interpretation of SHAP highlights features like Polyuria and offers insights into the model’s decision-making process. When compared to earlier studies, the proposed approach performs better at predicting diabetes in its early stages. The study shows that the model of deep learning and XAI could enhance the accuracy of diabetes prediction that can benefit the patient management and health care system decisions every day.