EEG-Based Brain Age Modeling in African Populations: Quantifying Cross-Population Generalization Gap and Brain Age Gap in Healthy Nigerian Adults Using Low-Cost EEG

Brain age modeling estimates biological brain aging from neurophysiological features, but existing models lack validation in African populations. We developed a resting-state electroencephalogram (EEG) framework using an international training corpus of 1,951 healthy recordings from TUAB (N=1,344) and the Dortmund Vital Study (N=607), and evaluated its transferability to 77 healthy Nigerian controls from the ASZED cohort. Continuous recordings were harmonized to a standard 19-channel 10–20 montage to extract a 190-dimensional feature space including spectral power, oscillatory pacing, cross-frequency ratios, complexity, and aperiodic activity. Gradient Boosting Regressors (GBR) and Random Forest (RF) models were trained via 10-fold cross-validation on the international corpus, tested under direct zero-shot transfer, and adapted using a few-shot approach where the target Nigerian cohort was partitioned into a 25-subject calibration subset and an independent 52-subject held-out test cohort. GBR achieved an mean absolute error (MAE) of 10.54 years (R2 = 0.27), and RF achieved 10.85 years (R2 = 0.25). Direct zero-shot transfer yielded MAEs of 11.04 years (GBR) and 11.32 years (RF), showing systematic positive biases of +4.17 and +6.33 years. Few-shot calibration on the 25 local subjects reduced prediction error on the independent 52-subject test cohort from 10.98 to 9.99 years for RF (a 0.98-year recovery) and from 9.80 to 9.43 years for GBR, while recentering the held-out Brain Age Gap near zero (−0.09 ± 3.47 years for RF; −1.44 ± 7.20 years for GBR). These findings demonstrate that lightweight post-hoc calibration on small local samples can recover cross-population transportability, providing an initial normative electrophysiological benchmark for low-resource clinical settings.