Sampaio, TatianaEncarnação, SamuelAmaro, BrunaRibeiro, JoanaBranquinho, LuísMonteiro, António M.Teixeira, José EduardoHattabi, SoukainaSortwell, AndrewLeite, Luciano BernardesMalheiro, AlexandraRodrigues, Pedro M.Beat KnechtleFlores, Pedro MiguelForte, Pedro2026-02-252026-02-252026Sampaio, Tatiana; Encarnação, Samuel; Amaro, Bruna; Ribeiro, Joana; Branquinho, Luís; Monteiro, António M.; Teixeira, José Eduardo; Hattabi, Soukaina; Sortwell, Andrew; Leite, Luciano Bernardes; Malheiro, Alexandra; Rodrigues, Pedro M.; Knechtle, Beat; Flores, Pedro Miguel; Forte, Pedro (2026). Machine learning prediction of adolescent obesity using physical fitness data. Obesity Medicine. ISSN 2451-8476. 59, p. 1-92451-8476http://hdl.handle.net/10198/35866The escalating prevalence of obesity among adolescents has emerged as a critical global public health challenge. Machine learning techniques have been used to predict obesity in adolescents. This study aimed to develop and validate a robust obesity prediction model for adolescents using this hybrid approach, leveraging data from a diverse cross-sectional population-based study. The hybrid method combines statistical inference with non-linear machine learning to enhance prediction accuracy. Physical fitness data were collected from the FITescola® tests. Multiple tests were employed to evaluate physical fitness. Multiple Poisson's multiple regression method was applied to identify the most predictive variables set of the adolescent's body mass index (BMI) classification. The model's goodness-of-fit statistics indicate a strong fit, with a log-likelihood of 􀀀 8068.6 and a Pseudo R-squared value of 0.8853, where the aerobic fitness (AF), upper limb strength (ULS) and lower limb flexibility (LLF) presented an inverse association with the adolescent's BMI. In contrast the adolescent's core strength presented a positive association with their body mass. The random forest regression showed that an average of 35 repetition on the yo-yo test predicted a healthy BMI percentile [predBMIperc = 0.31]. In addition, the model presented good validity [MAE = 0.36, MSE = 0.20, RMSE = 0.45, R2 = 0.54]. The model's strong fit and accurate random forest regression's predictions suggest that physical fitness components, such as aerobic fitness, upper limb strength, lower limb power, and core strength, play a significant role in obesity risk among adolescents.engjournal article10.1016/j.obmed.2025.100679