Heart rate variability (HRV) reflects the autonomic regulation of cardiac activity and serves as an important biomarker in clinical and physiological research, providing insights into stress, fatigue, and cardiovascular health. Traditionally, accurate HRV estimation relies on electrocardiogram (ECG) recordings, which require precise electrode placement and controlled measurement conditions. Although photoplethysmography (PPG) signals can also be used to estimate HRV, their accuracy is often reported to be suboptimal. In this work, we demonstrate that HRV estimation from PPG signals can be significantly improved by leveraging lightweight machine learning models. Our results show that the proposed approach achieves substantial accuracy improvements over widely used PPG-based HRV calculation methods, while remaining well within the resource constraints of wearable IoT devices.