Behavioral and Psychological Symptoms of Dementia (BPSD) significantly impact dementia care, presenting challenges for both patients and caregivers. Recognizing the occurrence of BPSD at an early stage and effectively managing BPSD is critical for alleviating caregiver burden and reducing strain on healthcare systems. While utilizing machine learning technologies has enhanced the performance of BPSD prediction, there is still a lack of an integrated system for BPSD prediction and alerting. In this study, a smart wearable device-based BPSD Alerting System was developed to present a real-world clinical application. This system can send alert notifications to users when the predicted BPSD probability is higher than the user-specific threshold. We also introduced two BPSD prediction studies and their methodologies. Besides, we offered insights into their approaches to personalized and generalized modeling and their binary and ternary (fine-grained) classification strategies. This review highlights the strengths and limitations of each framework and provides valuable guidance for developing an effective BPSD Alerting System. To the best of our knowledge, this is the first work to mention the integrated system based on smart wearable devices to implement machine learning applications of BPSD Alerting in a clinical environment, laying the foundation for further innovations in personalized dementia care in geriatric healthcare.