This study evaluated whether frailty among older adults living in long-term care (LTC) facilities can be accurately identified using data from a single wearable 3-D accelerometer that captures both short-duration gait tasks and approximately one week of daily physical activity. Fifty-one residents, average age about 85 years, completed the assessment, and researchers extracted 34 gait features, 3 activity variables, and 6 demographic factors. Five machine-learning models were tested using leave-one-out cross-validation, with the Extreme Gradient Boosting model performing best, achieving an accuracy of 86.3% and an AUC of 0.92. Explainable AI techniques showed that participants classified as frail tended to have more variable, complex, and asymmetric gait patterns, such as increased stride-length variability, higher sample entropy, and greater gait asymmetry, compared with non-frail residents.