This paper presents SmartFall, an Android app that uses accelerometer data collected from a commodity-based smartwatch Internet of Things (IoT) device to detect falls. The smartwatch is paired with a smartphone that runs the SmartFall application, which performs the computation necessary for the prediction of falls in real time without incurring latency in communicating with a cloud server, while also preserving data privacy. We experimented with both traditional (Support Vector Machine and Naive Bayes) and non-traditional (Deep Learning) machine learning algorithms for the creation of fall detection models using three different fall datasets (Smartwatch, Notch, Farseeing). Our results show that a Deep Learning model for fall detection generally outperforms more traditional models across the three datasets. This is attributed to the Deep Learning model’s ability to automatically learn subtle features from the raw accelerometer data that are not available to Naive Bayes and Support Vector Machine, which are restricted to learning from a small set of extracted features manually specified. Furthermore, the Deep Learning model exhibits a better ability to generalize to new users when predicting falls, an important quality of any model that is to be successful in the real world. We also present a three-layer open IoT system architecture used in SmartFall, which can be easily adapted for the collection and analysis of other sensor data modalities (e.g., heart rate, skin temperature, walking patterns) that enables remote monitoring of a subject’s wellbeing.

This article presents an in-depth experimental study of Ensemble Deep Learning techniques on small datasets for the analysis of time-series data generated by wearable devices. Deep Learning networks generally require large datasets for training. In some health care applications, such as the real-time smartwatch-based fall detection, there are no publicly available, large, annotated datasets that can be used for training, due to the nature of the problem (i.e., a fall is not a common event). We conducted a series of offline experiments using two different datasets of simulated falls for training various ensemble models. Our offline experimental results show that an ensemble of Recurrent Neural Network (RNN) models, combined by the stacking ensemble technique, outperforms a single RNN model trained on the same data samples. Nonetheless, fall detection models trained on simulated falls and activities of daily living performed by test subjects in a controlled environment, suffer from low precision due to high false-positive rates. In this work, through a set of real-world experiments, we demonstrate that the low precision can be mitigated via the collection of false-positive feedback by the end-users. The final Ensemble RNN model, after re-training with real-world user archived data and feedback, achieved a significantly higher precision without reducing much of the recall in a real-world setting.