BACKGROUND The accuracy of movement determination software of currently available activity trackers is not only insufficient for scientific applications, but also not open source. We developed an accurate, open-source, smartphone-based activity tracking toolbox, consisting of an Android app and two different deep learning algorithms adaptable to new behaviors. OBJECTIVE To offer an open source, adaptable machine learning toolbox for movement recognition, which can bes trained to specific needs and can yield repeatable results across multiple studies. The main focus lies on comprehensibility and traceability of the classification, which is not provided by the most widely used software. METHODS Using a semi-supervised deep learning approach, we identify different classes of activity, based on accelerometry and gyroscopy data, based on own and open-competition data. RESULTS With robustness against variation in sampling rate and sensor dimensional input, we achieved ~87% accuracy in classifying 6 different behaviors on own data and MotionSense Data. Tested on own data, the accuracy drops to 26%, which shows superiority of our own algorithm. CONCLUSIONS Human Activity Recorder is a versatile, retrainable toolbox, open-source available and accurate, which is tested on new data continually, enables researchers to adapt to the behavior measured and achieve repeatability in science.

Background The accuracy of movement determination software in current activity trackers is insufficient for scientific applications, which are also not open-source. Objective To address this issue, we developed an accurate, trainable, and open-source smartphone-based activity-tracking toolbox that consists of an Android app (HumanActivityRecorder) and 2 different deep learning algorithms that can be adapted to new behaviors. Methods We employed a semisupervised deep learning approach to identify the different classes of activity based on accelerometry and gyroscope data, using both our own data and open competition data. Results Our approach is robust against variation in sampling rate and sensor dimensional input and achieved an accuracy of around 87% in classifying 6 different behaviors on both our own recorded data and the MotionSense data. However, if the dimension-adaptive neural architecture model is tested on our own data, the accuracy drops to 26%, which demonstrates the superiority of our algorithm, which performs at 63% on the MotionSense data used to train the dimension-adaptive neural architecture model. Conclusions HumanActivityRecorder is a versatile, retrainable, open-source, and accurate toolbox that is continually tested on new data. This enables researchers to adapt to the behavior being measured and achieve repeatability in scientific studies.