This paper focuses on forecasting minute-by-minute stress, anxiety, and affective states using wearable sensor data. It addresses mental health as a growing concern and the limitations of traditional assessment methods. A time-series machine learning framework was developed using electrodermal activity (EDA) and heart rate variability (HRV) features from the WESAD dataset. Models were trained and evaluated for minute-by-minute prediction of self-reported psychological states. Both classification (stress, anxiety) and regression models (affect) were explored comparing time-series and static approaches. Findings support the feasibility of real-time, personalized mental health monitoring using wearable devices and their potential for timely interventions in clinical or remote settings. The paper demonstrates how temporal modeling can enhance emotional state prediction and inform future research and system development.