Diabetes mellitus is linked to vascular complications, including arterial stiffness and endothelial dysfunction, which serve as key biomarkers for early detection. This study investigates a non-invasive approach to assess these impairments using photoplethysmography (PPG) signals from healthy and diabetic individuals. The acquired signals were preprocessed followed by feature extraction, yielding 620 features. Feature selection techniques—ReliefF and Minimum Redundancy Maximum Relevance (MRMR)—reduced the dataset to 20 key features. Statistical analysis was conducted to identify significant differences between groups, and machine learning models, K-Nearest Neighbors (KNN) and Support Vector Machine (SVM), were employed for classification. The optimized KNN model achieved 74% accuracy with all features and 72.7% with the selected features, while SVM attained 73.3% and 67.3%, respectively. These findings highlight the potential of integrating statistical and machine learning techniques for non-invasive diabetes detection through vascular health assessment.