Insufficient physical activity presents a significant hazard to overall health, with sedentary lifestyles linked to a variety of health issues. Monitoring physical activity levels allows the recognition of patterns of sedentary behavior and the provision of coaching to meet the recommended physical activity standards. In this paper, we aim to address the problem of reducing the time consuming process of fitting classifiers when generating personalized models for a coaching application. The proposed approach consists of evaluating the effects of clustering participants based on their walking patterns and then recommending a unique model for each group. Each model consists of a random forest classifier with a different number of estimators each. The resulting approach reduces the fitting time considerably while keeping nearly the same classification performance as personalized models.
DOCUMENT
We present a novel architecture for an AI system that allows a priori knowledge to combine with deep learning. In traditional neural networks, all available data is pooled at the input layer. Our alternative neural network is constructed so that partial representations (invariants) are learned in the intermediate layers, which can then be combined with a priori knowledge or with other predictive analyses of the same data. This leads to smaller training datasets due to more efficient learning. In addition, because this architecture allows inclusion of a priori knowledge and interpretable predictive models, the interpretability of the entire system increases while the data can still be used in a black box neural network. Our system makes use of networks of neurons rather than single neurons to enable the representation of approximations (invariants) of the output.
LINK
In wheelchair sports, there is an increasing need to monitor mechanical power in the field. When rolling resistance is known, inertial measurement units (IMUs) can be used to determine mechanical power. However, upper body (i.e., trunk) motion affects the mass distribution between the small front and large rear wheels, thus affecting rolling resistance. Therefore, drag tests – which are commonly used to estimate rolling resistance – may not be valid. The aim of this study was to investigate the influence of trunk motion on mechanical power estimates in hand-rim wheelchair propulsion by comparing instantaneous resistance-based power loss with drag test-based power loss. Experiments were performed with no, moderate and full trunk motion during wheelchair propulsion. During these experiments, power loss was determined based on 1) the instantaneous rolling resistance and 2) based on the rolling resistance determined from drag tests (thus neglecting the effects of trunk motion). Results showed that power loss values of the two methods were similar when no trunk motion was present (mean difference [MD] of 0.6 1.6 %). However, drag test-based power loss was underestimated up to −3.3 2.3 % MD when the extent of trunk motion increased (r = 0.85). To conclude, during wheelchair propulsion with active trunk motion, neglecting the effects of trunk motion leads to an underestimated mechanical power of 1 to 6 % when it is estimated with drag test values. Depending on the required accuracy and the amount of trunk motion in the target group, the influence of trunk motion on power estimates should be corrected for.
DOCUMENT
