Diabetes prediction is a crucial healthcare challenge, where accurate and efficient classification models play a pivotal role in early diagnosis and management. This study presents a comprehensive approach combining traditional machine learning, quantum-enhanced algorithms, and data balancing techniques to improve predictive accuracy for diabetes detection. Initially, the Adaptive Synthetic (ADASYN) oversampling method is applied to address class imbalance in the dataset, ensuring equitable model training. Traditional machine learning models, including Random Forest, Gradient Boosting, Naive Bayes, SVM, and AdaBoost, are trained on the balanced dataset. These models are evaluated using key metrics such as accuracy, precision, recall, F1-score, with Random Forest achieving the best performance among traditional models, with an accuracy of 80% and an ROC-AUC of 0.89. Subsequently, quantum machine learning techniques, including quantum support vector classifier (QSVC), quantum k-nearest neighbors (Qk-NN), and quantum neural networks (QNN), are employed. The results demonstrate the significant potential of quantum-enhanced models, with QSVC achieving 90% accuracy, Qk-NN achieving 85% accuracy, and QNN surpassing others with 92% accuracy. This hybrid approach, augmented with ADASYN for improved data balance, underscores the superiority of quantum models in leveraging advanced computation to achieve higher predictive performance. The findings highlight the promise of integrating quantum computing into healthcare analytics, paving the way for more accurate diabetes prediction.