Abstract:Accelerometers are widely used to measure physical activity behaviour, including in children. The traditional method for processing acceleration data uses cut points to define physical activity intensity, relying on calibration studies that relate the magnitude of acceleration to energy expenditure. However, these relationships do not generalise across diverse populations and hence they must be parametrised for each subpopulation (e.g., age groups) which is costly and makes studies across diverse populations and over time difficult. A data-driven approach that allows physical activity intensity states to emerge from the data, without relying on parameters derived from external populations, offers a new perspective on this problem and potentially improved results. We applied an unsupervised machine learning approach, namely a hidden semi-Markov model, to segment and cluster the raw accelerometer data recorded (using a waist-worn ActiGraph GT3X+) from 279 children (9–38 months old) with a diverse range of developmental abilities (measured using the Paediatric Evaluation of Disability Inventory–Computer Adaptive Testing measure). We benchmarked this analysis with the cut points approach, calculated using thresholds from the literature which had been validated using the same device and for a population which most closely matched ours. Time spent active as measured by this unsupervised approach correlated more strongly with PEDI-CAT measures of the child's mobility (R 2 : 0.51 vs 0.39), social-cognitive capacity (R 2 : 0.32 vs 0.20), responsibility (R 2 : 0.21 vs 0.13), daily activity (R 2 : 0.35 vs 0.24), and age (R 2 : 0.15 vs 0.1) than that measured using the cut points approach. Unsupervised machine learning offers the potential to provide a more sensitive, appropriate, and cost-effective approach to quantifying physical activity behaviour in diverse populations, compared to the current cut points approach. This, in turn, supports research that is more inclusive of diverse or rapidly changing populations. Physical activity participation in young children has often been measured using parent reports. Accelerometry provides a more objective measurement but the traditional methods used to quantify this require calibration and struggle to generalise to diverse or rapidly changing populations such as young children. In recent years unsupervised machine learning methods have been shown to be able to segment and cluster accelerometry, allowing categories of activity intensity to emerge from a data-driven process. Here we show that an unsupervised machine learning technique (the hidden semi-Markov model) can be used to estimate categories of activity intensity in accelerometry data recorded from a diverse population of children age 9–36 months. We also show that this approach better captures the variance of movement abilities in the population than the traditional cut points approach. The hidden semi-Markov model approach provides a more effective approach for processing and analysing accelerometer data in rapidly changing and diverse populations such as young children, compared to the more traditional cut points approach. As it does not require calibration studies to incorporate new populations it has the potential to facilitate inclusion of unrepresented populations in research, as well as being less resource intensive.

Development of a multi-wear-site, deep learning-based physical activity intensity classification algorithm using raw acceleration data

Development of machine learning-based algorithms for classifying physical activity intensity using wrist and thigh worn wearables

Development and validation of a smartwatch algorithm for differentiating physical activity intensity in health monitoring

Classifying Tasks Performed by Electrical Line Workers Using a Wrist-Worn Sensor: A Data Analytic Approach

Deep learning-based classification with improved time resolution for physical activities of children

Analysis of Personal Life Patterns Using Accelerometer-Based Wearable Devices

Machine Learning on Prediction of Relative Physical Activity Intensity Using Medical Radar Sensor and 3D Accelerometer

Using machine learning methods to predict physical activity types with Apple Watch and Fitbit data using indirect calorimetry as the criterion.

Accurate Step Count with Generalized and Personalized Deep Learning on Accelerometer Data

Using unsupervised machine learning to quantify physical activity from accelerometry in a diverse and rapidly changing population

Development of cut-points for determining activity intensity from a wrist-worn ActiGraph accelerometer in free-living adults

0280 Unlocking the Potential of Consumer Wearables for Predicting Sleep in Children: A Device-Agnostic Machine Learning Approach

Prediction of pediatric activity intensity with wearable sensors and bi-directional LSTM models

Sedentary Behavior Estimation with Hip-worn Accelerometer Data: Segmentation, Classification and Thresholding

Sensor Positioning for Activity Recognition Using Wearable Accelerometers.

Daily Physical Activity Monitoring -- Adaptive Learning from Multi-source Motion Sensor Data

Energy Expenditure Prediction from Accelerometry Data Using Long Short-Term Memory Recurrent Neural Networks

Intensity Thresholds on Raw Acceleration Data: Euclidean Norm Minus One (ENMO) and Mean Amplitude Deviation (MAD) Approaches

Sensor Placement for Activity Detection Using Wearable Accelerometers

Real-time Human Activity Classification by Accelerometer Embedded Wearable Devices.

Calibration and Cross-Validation of Accelerometery for Estimating Movement Skills in Children Aged 8–12 Years