Stress and anxiety in university students are major concerns for both mental health and academic performance. This work presents a physiological monitoring approach for the early identification of individualized physiological deviations potentially associated with stress and anxiety, based on data collected through the Fitbit Inspire 3 from 38 undergraduate and graduate students. The analyzed variables included heart rate (HR), heart rate variability (HRV), peripheral oxygen saturation (SpO2), respiratory rate, sleep, and physical activity. The methodology included signal selection and preprocessing, data anonymization, the construction of a common 5-min time grid, and intra-individual analysis using personalized baselines, percentiles, and sliding windows. Based on this approach, early alerts were generated for each student, and their persistence was evaluated over intraday and daily horizons. Additionally, a predictive stage was incorporated using an ARIMA (Autoregressive Integrated Moving Average) model and a Multilayer Perceptron (MLP) neural network, with the 7-day risk series as the target variable. The comparison between both approaches was carried out through temporal validation, reserving the last month of data as the test set. Results show that ARIMA achieved lower prediction error and higher alert accuracy compared to the MLP. Overall, the study demonstrates the feasibility of an early alert system based on wearable data and individualized temporal analysis. The generated risk score should be interpreted as a physiological deviation indicator, not as a clinical or psychometric diagnosis of stress or anxiety, with potential use as a support tool for health professionals in the preventive monitoring of students¿ mental well-being.
