Mobile health applications are considered to be powerful tools for activity-based wellness management. With the availability of multimodal sensors in smart devices used in our daily lives, it is possible to track human activity and deliver context-aware wellness services. The embedded sensors in naturally used devices such as smartphones, smartwatches, and wearables contain rich information that can be integrated for human activity recognition. Our research demonstrates how powerful boosting algorithms can extract knowledge for human activity classification in a real-life setting. Our results show that boosting classifiers outperform traditional machine learning classifiers in the detection of basic human activities such as walking, standing, sitting, exercise, and sleeping. Further, we perform feature engineering to compare the potential of a smartphone and a smartwatch in activity detection. Our feature engineering strategy provides directions about the selection of sensor features for improvement in classification of basic human activities. The theoretical and practical implications of this research for activity-based wellness management are also discussed.

移动健康应用程序被认为是基于活动的健康管理的强大工具。随着我们日常生活中使用的智能设备中多模式传感器的可用性，可以跟踪人类活动并提供情境感知的健康服务。智能手机，智能手表和可穿戴设备等自然使用的设备中的嵌入式传感器包含丰富的信息，可以集成这些信息以识别人类活动。我们的研究表明，强大的增强算法如何在现实环境中提取人类活动分类的知识。我们的结果表明，在检测基本人类活动（例如步行，站立，坐下，运动和睡眠）时，增强分类器优于传统机器学习分类器。进一步，我们进行功能工程设计，以比较智能手机和智能手表在活动检测中的潜力。我们的特征工程策略为选择传感器特征提供了指导，以改善人类基本活动的分类。还讨论了这项研究对基于活动的健康管理的理论和实践意义。