Human Activity Recognition (HAR) has become a major focus in research, primarily due to its vital role in security applications, healthcare, entertainment, and other domains. At the core of many intelligent systems, which focus on human needs and activities, the recognition of such activities remains a challenging yet researchable area. The recognition algorithms require substantial amounts of data instances and labels for effective training. Many approaches have been designed to identify human activities by utilizing the relationships among multiple labels and instances generated by IoT technology within smart home environments. This article reviews existing multi-label learning approaches and the research conducted to recognize multiple human activities through ambient IoT sensors in smart homes. The study reveals that using multi-label learning for activity recognition is more effective than traditional machine learning approaches, especially for handling the complexities of multi-resident activities generated by various IoT sensors in smart homes. Several challenges need to be addressed to enhance the recognition of multi-resident activities, such as complexity, high dimensionality, association, heterogeneity, and feature extraction techniques. While different researchers have addressed most of these issues to some extent, there remains a need for greater focus due to the importance of these challenges in assisting control systems that provide services to caregivers, the elderly, vulnerable individuals, and medical and security personnel. Addressing these challenges more comprehensively will significantly improve the efficiency and usefulness of smart environments.