Continuous, non-invasive blood pressure monitoring represents a critical unmet need in cardiovascular healthcare. Conventional oscillometric methods provide only intermittent readings and impose practical constraints that limit their utility in ambulatory and long-term monitoring scenarios. This work presents a machine learning framework for cuffless blood pressure estimation grounded in the physiological relationship between arterial stiffness and pulse wave propagation, here captured through a differential Pulse Transit Time (dPTT) — defined as the time delay between two peripheral pulse waveforms recorded at the finger and toe sites, in the absence of a proximal (e.g., ECG) reference. A comprehensive feature set is constructed from the dPTT signal, encompassing inverse dPTT, logarithmic transformations, pulse wave reflection indices, heart rate interaction terms, pulse wave velocity, and age-stratified vascular stiffness descriptors. These features are designed to jointly capture the elastic and hemodynamic properties of the arterial wall across a physiologically diverse population. A Random Forest regression model is trained on 500 subjects drawn from the Pulse Wave Database (PWDB) and evaluated under a leave-one-out cross-validation protocol to ensure unbiased generalization assessment. The proposed framework achieves a mean absolute error of 6.33 mmHg for systolic blood pressure (SBP) and 4.01 mmHg for diastolic blood pressure (DBP), with corresponding R² values of 0.719 and 0.586. These results satisfy the British Hypertension Society (BHS) grading criteria and demonstrate strong agreement between model predictions and reference values across the evaluated cohort. These results demonstrate that physiologically informed feature engineering combined with ensemble learning can achieve accurate blood pressure estimation using peripheral pulse propagation signals. This approach provides a robust basis for developing wearable, cuffless cardiovascular monitoring systems for continuous assessment in clinical and real-world environments.