Accurate longitudinal velocity estimation is essential for vehicle control systems such as Electronic Stability Control, Anti-lock Braking Systems, and torque vectoring. In production vehicles, this estimation typically relies on wheel speed sensors and inertial measurement units, sometimes supplemented by Global Navigation Satellite System measurements. However, these sensors present inherent limitations: integration-based drift in inertial measurements, unreliable wheel speed readings under high-slip conditions, and degraded satellite signals in obstructed environments. Sensor fusion techniques are employed to combine information from multiple sensors and mitigate their individual limitations. While model-based algorithms, particularly Kalman filtering, are widely adopted, increasing data availability and computational capabilities have motivated the exploration of data-driven techniques as alternative or complementary approaches to sensor fusion and state estimation.