This study proposes a novel deep learning (DL) technique for early diabetes prediction with high accuracy. The model uses a Deep Neural Network (DNN)-based Multi-Layer Perceptron (MLP) algorithm with ten hidden layers and multiple epochs to process data from the Mansoura University Children's Hospital Diabetes (MUCHD) dataset, which includes 548 patients with 18 features. The system was validated using cross-validation and various metrics such as accuracy, F-score, precision, sensitivity, specificity, and Dice similarity coefficient, achieving a remarkable accuracy of 99.8%. The proposed method outperforms existing state-of-the-art methods by 0.39% in overall system performance. Diabetes is a chronic disease caused by excessive glucose in the bloodstream, leading to complications such as visual impairment, cardiovascular issues, and nerve damage. Early detection is critical for timely treatment. DL, a subset of machine learning, uses supervised deep neural networks to process and classify large datasets, enabling early disease detection. The study compares DL techniques with traditional methods, highlighting the effectiveness of DL in diabetes prediction. The proposed system includes data collection, preprocessing, classification, training/test, evaluation, optimization, and prediction phases. The dataset includes features such as age, sex, duration, cholesterol, creatinine, HbA1c, insulin, and blood glucose levels. Missing values were handled by replacing them with mean values, and features were scaled to a normalized range. The model used ReLU and Sigmoid activation functions, with hyperparameters fine-tuned for optimal performance. The system achieved high accuracy in classification metrics such as precision, recall, F1-score, and specificity. The model demonstrated robust performance in various quality measures, including accuracy, training score, mean squared error, and R2 score. The study concludes that the proposed DL-based system is effective for predicting diabetes and can be applied to other binary disease classification systems. The system's adaptability and efficiency make it a valuable tool for early diabetes detection and management.This study proposes a novel deep learning (DL) technique for early diabetes prediction with high accuracy. The model uses a Deep Neural Network (DNN)-based Multi-Layer Perceptron (MLP) algorithm with ten hidden layers and multiple epochs to process data from the Mansoura University Children's Hospital Diabetes (MUCHD) dataset, which includes 548 patients with 18 features. The system was validated using cross-validation and various metrics such as accuracy, F-score, precision, sensitivity, specificity, and Dice similarity coefficient, achieving a remarkable accuracy of 99.8%. The proposed method outperforms existing state-of-the-art methods by 0.39% in overall system performance. Diabetes is a chronic disease caused by excessive glucose in the bloodstream, leading to complications such as visual impairment, cardiovascular issues, and nerve damage. Early detection is critical for timely treatment. DL, a subset of machine learning, uses supervised deep neural networks to process and classify large datasets, enabling early disease detection. The study compares DL techniques with traditional methods, highlighting the effectiveness of DL in diabetes prediction. The proposed system includes data collection, preprocessing, classification, training/test, evaluation, optimization, and prediction phases. The dataset includes features such as age, sex, duration, cholesterol, creatinine, HbA1c, insulin, and blood glucose levels. Missing values were handled by replacing them with mean values, and features were scaled to a normalized range. The model used ReLU and Sigmoid activation functions, with hyperparameters fine-tuned for optimal performance. The system achieved high accuracy in classification metrics such as precision, recall, F1-score, and specificity. The model demonstrated robust performance in various quality measures, including accuracy, training score, mean squared error, and R2 score. The study concludes that the proposed DL-based system is effective for predicting diabetes and can be applied to other binary disease classification systems. The system's adaptability and efficiency make it a valuable tool for early diabetes detection and management.