Continuous health monitoring using wearable devices has emerged as a promising approach to detecting early signs of physiological and daily activity anomalies in older adults living independently. However, traditional threshold-based methods often fail to account for individual variability, leading to excessive false positives or missed critical events. This study proposes an unsupervised anomaly detection framework that personalizes activity monitoring by clustering based on K-means to consider individual behavioral patterns. Using one-minute time window data from wearable sensors, we establish baseline activity profiles over seven days and detect deviations indicative of potential health risks. Four primary anomaly types were identified: nocturnal restlessness, exertion without movement, prolonged inactivity, and gradual reduction in activity. The system effectively localized these anomalies within a latent feature space using a non-linear dimensionality reduction technique, thus demonstrating its ability to differentiate between distinct deviation patterns.