With the increasing adoption of electrical vehicles (EV) by the general public, a lot of research is being conducted in Li-ion battery-related topics, having state-of-health (SoH) estimation a prominent role. In this work, machine-learning techniques are applied to estimate this parameter online in EV applications and in diverse scenarios. After a thorough analysis regarding cell ageing and the main factors influencing this process, a total of three approaches are developed: the first one is based on voltage measurements at fixed state-of-charge (SoC) levels, while the second one uses the charge gradients between certain voltage milestones as health indicators. The last method predicts capacity- and impedance-based SoH from a limited set of impedance measurements. The proposed approaches are tested for different chemistries and in various realistic scenarios to evaluate their performance and applicability. Great accuracy was obtained in all cases, with MAE as low as 0.4% when making future predictions, 0.5% when inferring at multiple temperatures, 0.4% for diverse, realistic ageing, and 0.2% for storage ageing. Thus, the methods constitute a powerful and viable alternative for online SoH estimation in real-world EV.