SleepLess: personalized sleep monitoring using smartphones and semi-supervised learning
SleepLess: personalized sleep monitoring using smartphones and semi-supervised learning
复制标题
DOI：
10.1007/s40012-023-00389-8
复制
发表时间：
2023-11
期刊：
影响因子：
--
通讯作者：
P. Mammen;Camellia Zakaria;Prashant J. Shenoy
中科院分区：
文献类型：
--
作者：
P. Mammen;Camellia Zakaria;Prashant J. Shenoy
睡眠影响我们的身体机能，对促进每个人的健康至关重要。为此，睡眠健康监测研究最近引起了人们的兴趣，包括将数据驱动的人工智能技术与可穿戴、智能手机和非接触式传感模式的移动健康相结合。无论如何，总的来说，之前的工作需要收集足够的真实数据来开发个性化和高度准确的睡眠预测模型。这一要求固有地提出了这样一个挑战，即当在没有标记数据的情况下推断新用户的睡眠时，这种模型表现不佳。在本文中，我们提出了esleeppless，它使用半监督学习管道，对从用户智能手机网络活动中感知的未标记数据进行学习，以开发个性化模型并检测他们夜间的睡眠时间。具体来说，它在现有用户集上使用预训练模型，为新用户的未标记数据生成伪标签，并通过微调选择性地选择伪标签来实现个性化。我们的irb批准的用户研究发现，sleepplessmodel在12-27分钟睡眠时间误差和18-25分钟清醒时间误差之间的准确率约为96%。与其他试图用更少的标记数据进行预测的方法进行比较发现，无眠算法同样产生了最好的性能。我们的研究表明，通过使用半监督方法，利用从用户智能手机的网络活动中提取的未标记数据，实现个性化睡眠预测模型的可行性。
Sleep affects our bodily functions and is critical in promoting every individual’s well-being. To that end, sleep health monitoring research has gained interest recently, including coupling data-driven AI techniques with mHealth adaptations of wearable, smartphone, and contactless-sensing modalities. Regardless, prior works, by and large, require gathering sufficient ground truth data to develop personalized and highly accurate sleep prediction models. This requirement inherently presents a challenge of such models underperforming when inferring sleep on new users without labeled data. In this paper, we proposeSleepLess, which uses a semi-supervised learning pipeline over unlabeled data sensed from the user’s smartphone network activity to develop personalized models and detect their sleep duration for the night. Specifically, it uses a pre-trained model on an existing set of users to produce pseudo labels for unlabeled data of a new user and achieves personalization by fine-tuning over selectively picking the pseudo labels. Our IRB-approved user study foundSleepLessmodel yielding around 96% accuracy, between 12–27 min of sleep time error and 18–25 min of wake time error. Comparison against other approaches that sought to predict with fewer labeled data foundSleepLess, similarly yielding best performance. Our study demonstrates the feasibility of achieving personalized sleep prediction models by utilizing unlabeled data extracted from network activity of users’ smartphones, using a semi-supervised approach.