Thermal Forecasting of Wireless Earbuds Using Time-Series Machine Learning: Implications for Prolonged Auditory Safety

The integration of machine-learning-assisted thermal forecasting into radiofrequency (RF) safety assessment represents a critical methodological advance for consumer wearable devices. This study applies time-series modelling—including autoregressive integrated moving average (ARIMA) and biophysical Pennes Bioheat Equation (BHE) frameworks—to infrared-thermographic temperature datasets collected during 120-minute trials of six commercially available Bluetooth earbud models in Warri, Nigeria (n = 50 participants, aged 12–35 years). RF power density, electric field, and magnetic flux density were measured using a GQ EMF-390 multi-field meter at 2.402–2.480 GHz. Estimated SAR values ranged from 0.00408 to 0.00718 W/kg—less than 0.4% of the ICNIRP 2 W/kg limit—while ear canal temperatures reached peak rises of 3.1–4.9°C during call mode. ARIMA models (R² = 0.93–0.97) accurately forecasted thermal plateaus and post-use cooling trajectories, with RMSE < 0.31°C. Budget-tier devices (Oraimo FreePods 3, Tecno Hipods H2) exhibited the highest EMF emissions and sustained thermal burdens, with plateau durations of 68–70 minutes. Teenagers reported significantly greater symptom severity than adults across all domains (p < .001), consistent with physiological vulnerability in younger users. These findings establish thermal forecasting as a viable safety tool and provide device-specific, age-stratified evidence supporting precautionary usage guidelines.