Using Deep Learning for Energy Expenditure Estimation with wearable sensors

Accurate estimation of Energy Expenditure (EE) in ambulatory settings is a key element in determining the causal relation between aspects of human behavior related to physical activity and health. We present a new methodology for activity-specific EE algorithms. The proposed methodology models activity clusters using specific parameters that capture differences in EE within a cluster, and combines these models with Metabolic Equivalents (METs) derived from the compendium of physical activities. We designed a protocol consisting of a wide set of sedentary, household, lifestyle and gym activities, and developed a new activity-specific EE algorithm applying the proposed methodology. The algorithm uses accelerometer (ACC) and heart rate (HR) data acquired by a single monitoring device, together with anthropometric variables, to predict EE. Our model recognizes six clusters of activities independent of the subject in 52.6 hours of recordings from 19 participants. Increases in EE estimation accuracy ranged from 18 to 31% compared to state of the art single and multi-sensor activity-specific methods.

Objective: Objectively and efficiently measuring physical activity is a common issue facing the fields of medicine, public health, education, and sports worldwide. In response to the problem of low accuracy in predicting energy consumption during human motion using accelerometers, a prediction model for asynchronous energy consumption in the human body is established through various algorithms, and the accuracy of the model is evaluated. The optimal energy consumption prediction model is selected to provide theoretical reference for selecting reasonable algorithms to predict energy consumption during human motion. Methods: A total of 100 subjects aged 18-30years participated in the study. Experimental data for all subjects are randomly divided into the modeling group (n = 70) and validation group (n = 30). Each participant wore a triaxial accelerometer, COSMED Quark pulmonary function tester (Quark PFT), and heart rate band at the same time, and completed the tasks of walking (speed range: 2km/h, 3km/h, 4km/h, 5km/h, and 6km/h) and running (speed range: 7km/h, 8km/h, and 9km/h) sequentially. The prediction models were built using accelerometer data as the independent variable and the metabolic equivalents (METs) as the dependent variable. To calculate the prediction accuracy of the models, root mean square error (RMSE) and bias were used, and the consistency of each prediction model was evaluated based on Bland-Altman analysis. Results: The linear equation, logarithmic equation, cubic equation, artificial neural network (ANN) model, and walking-and-running two-stage model were established. According to the validation results, our proposed walking-and-running two-stage model showed the smallest overall EE prediction error (RMSE = 0.76METs, Bias = 0.02METs) and the best performance in Bland-Altman analysis. Additionally, it had the lowest error in predicting EE during walking (RMSE = 0.66METs, Bias = 0.03METs) and running (RMSE = 0.90METs, Bias < 0.01METs) separately, as well as high accuracy in predicting EE at each single speed. Conclusion: The ANN-based walking-and-running two-stage model established by separating walking and running can better estimate the walking and running EE, the improvement of energy consumption prediction accuracy will be conducive to more accurate to monitor the energy consumption of PA.

Our aim was to compare three research-grade accelerometers for their accuracy in step detection and energy expenditure (EE) estimation in a laboratory setting, at different speeds, especially in overweight/obese participants. Forty-eight overweight/obese subjects participated. Participants performed an exercise routine on a treadmill with six different speeds (1.5, 3, 4.5, 6, 7.5, and 9 km/h) for 4 min each. The exercise was recorded on video and subjects wore three accelerometers during the exercise: Sartorio Xelometer (SX, hip), activPAL (AP, thigh), and ActiGraph GT3X (AG, hip), and energy expenditure (EE) was estimated using indirect calorimetry for comparisons. For step detection, speed-wise mean absolute percentage errors for the SX ranged between 9.73–2.26, 6.39–0.95 for the AP, and 88.69–2.63 for the AG. The activPALs step detection was the most accurate. For EE estimation, the ranges were 21.41–15.15 for the SX, 57.38–12.36 for the AP, and 59.45–28.92 for the AG. All EE estimation errors were due to underestimation. All three devices were accurate in detecting steps when speed exceeded 4 km/h and inaccurate in EE estimation regardless of speed. Our results will guide users to recognize the differences, weaknesses, and strengths of the accelerometer devices and their algorithms.

Objectives: This study aimed to provide an improved energy expenditure estimation for heavy-load physical labor using accelerometer data and heart rate (HR) measured by wearables and to support food preparation and supply management for disaster relief and rescue operations as an expedition team. Methods: To achieve an individually optimized estimation for energy expenditure, a model equation parameter was determined based on the measurements of physical activity and HR during simulated rescue operations. The metabolic equivalent of task (MET), which was measured by using a tri-axial accelerometer and individual HR, was used, where two (minimum and maximum) or three (minimum, intermediate, and maximum) representative reference points were selected for each individual model fitting. In demonstrating the applicability of our approach in a realistic situation, accelerometer-based METs and HR of 30 males were measured using the tri-axial accelerometer and wearable HR during simulated rescue operations over 2days. Results: Data sets of 27 rescue operations (age:34.2 ± 7.5years; body mass index (BMI):22.9 ± 1.5kg/m2) were used for the energy expenditure estimation after excluding three rescue workers due to their activity type and insufficient HR measurement. Using the combined approach with a tri-axial accelerometer and HR, the total energy expenditure increased by 143% for two points and 133% for three points, compared with the estimated total energy expenditure using only the accelerometer-based method. Conclusion: The use of wearables provided a reasonable estimation of energy expenditure for physical workers with heavy equipment. The application of our approach to disaster relief and rescue operations can provide important insights into nutrition and healthcare management.

Accurate estimation of energy expenditure and gas exchange dynamics is essential for health monitoring and performance optimization. This study addresses the limitations of traditional systems by developing a portable, non-invasive, and real-time solution that correlates physiological signals with energy metabolism. The proposed system estimates energy expenditure and metabolic state using oxygen and carbon dioxide flows derived from non-invasive variables such as respiratory ventilation and heart rate. It utilizes Bluetooth Low Energy (BLE) for wireless communication and includes user-friendly interfaces for smartphones and computers to facilitate data visualization and recording. Calibration is performed using a calorimeter, resulting in an average estimation error of 14.83%. The system demonstrates reliable performance under various conditions, providing real-time estimations of energy expenditure and gas exchange. Its portability and ergonomic design improve usability; however, precise calibration remains essential, and broader testing is required to validate robustness. A key advantage of the system is its ability to operate entirely offline, relying solely on BLE for data transmission, making it suitable for real-time monitoring in diverse environments.

BackgroundEstimating energy expenditure with indirect calorimetry requires expensive equipment and several minutes of data collection for each condition of interest. While several methods estimate energy expenditure using correlation to data from wearable sensors, such as heart rate monitors or accelerometers, their accuracy has not been evaluated for activity conditions or subjects not included in the correlation process. The goal of our study was to develop data-driven models to estimate energy expenditure at intervals of approximately one second and demonstrate their ability to predict energetic cost for new conditions and subjects. Model inputs were muscle activity and vertical ground reaction forces, which are measurable by wearable electromyography electrodes and pressure sensing insoles.MethodsWe developed models that estimated energy expenditure while walking (1) with ankle exoskeleton assistance and (2) while carrying various loads and walking on inclines. Estimates were made each gait cycle or four second interval. We evaluated the performance of the models for three use cases. The first estimated energy expenditure (in Watts) during walking conditions for subjects with some subject specific training data available. The second estimated all conditions in the dataset for a new subject not included in the training data. The third estimated new conditions for a new subject.ResultsThe mean absolute percent errors in estimated energy expenditure during assisted walking conditions were 4.4%, 8.0%, and 8.1% for the three use cases, respectively. The average errors in energy expenditure estimation during inclined and loaded walking conditions were 6.1%, 9.7%, and 11.7% for the three use cases. For models not using subject-specific data, we evaluated the ability to order the magnitude of energy expenditure across conditions. The average percentage of correctly ordered conditions was 63% for assisted walking and 87% for incline and loaded walking.ConclusionsWe have determined the accuracy of estimating energy expenditure with data-driven models that rely on ground reaction forces and muscle activity for three use cases. For experimental use cases where the accuracy of a data-driven model is sufficient and similar training data is available, standard indirect calorimetry could be replaced. The models, code, and datasets are provided for reproduction and extension of our results.

SMETool: A web-based tool for soil moisture estimation based on Eo-Learn framework and Machine Learning methods

The aim of this study was to compare the energy expenditure (EE) estimations of activity-specific prediction equations (ASPE) and of an artificial neural network (ANNEE) based on accelerometry with measured EE. Forty-three children (age: 9.8 ± 2.4 yr) performed eight different activities. They were equipped with one tri-axial accelerometer that collected data in 1-s epochs and a portable gas analyzer. The ASPE and the ANNEE were trained to estimate the EE by including accelerometry, age, gender, and weight of the participants. To provide the activity-specific information, a decision tree was trained to recognize the type of activity through accelerometer data. The ASPE were applied to the activity-type-specific data recognized by the tree (Tree-ASPE). The Tree-ASPE precisely estimated the EE of all activities except cycling [bias: -1.13 ± 1.33 metabolic equivalent (MET)] and walking (bias: 0.29 ± 0.64 MET; P < 0.05). The ANNEE overestimated the EE of stationary activities (bias: 0.31 ± 0.47 MET) and walking (bias: 0.61 ± 0.72 MET) and underestimated the EE of cycling (bias: -0.90 ± 1.18 MET; P < 0.05). Biases of EE in stationary activities (ANNEE: 0.31 ± 0.47 MET, Tree-ASPE: 0.08 ± 0.21 MET) and walking (ANNEE 0.61 ± 0.72 MET, Tree-ASPE: 0.29 ± 0.64 MET) were significantly smaller in the Tree-ASPE than in the ANNEE (P < 0.05). The Tree-ASPE was more precise in estimating the EE than the ANNEE. The use of activity-type-specific information for subsequent EE prediction equations might be a promising approach for future studies.

Exploring the optimization of communication strategies for animation films in the context of cross-cultural communication, this research integrates the Internet of Things (IoT) and convolutional networks. The research constructs a collaborative filtering (CF) movie recommendation model based on a graph convolutional neural network (GCN) and investigates its application in cross-cultural communication. The fusion of IoT and convolutional networks in movie communication is also analyzed, and the effectiveness of the proposed GCN-CF model is validated through comparative experiments. The results indicate that, compared to other models, the GCN-CF model achieves the lowest Root Mean Square Error (RMSE) on the MovieLens 100 K and MovieLens 1 M datasets, with values of 0.8762 and 0.8275, respectively. Compared to traditional models, the GCN-CF model exhibits significantly superior performance in terms of RMSE, with reductions ranging from 0.6 to 5.2%, highlighting its heightened detection accuracy and overall performance. Moreover, the performance of the GCN-CF model is enhanced after introducing attention mechanisms and auxiliary information on both datasets, showing an improvement of 0.4% compared to the scenario without these additions. This data demonstrates the effectiveness of attention mechanisms and auxiliary information. Finally, the research presents an animation film communication strategy based on IoT and convolutional networks, offering novel insights for film production and communication, along with positive implications for cultural exchange and the advancement of the global media industry.

This paper develops a new predictor of walking energy expenditure from wireless measurements of body movements using triaxial accelerometers. Reliable data were collected from repeated walking experiments in different conditions on a treadmill with simultaneous measurement of expired oxygen and carbon dioxide. Support vector regression, a powerful non-linear regression method, was used to process and model the data. This novel processing method sets this investigation apart from existing papers. Good results were achieved in the robust estimation of walking related energy expenditure from a number of variables derived from triaxial accelerometer and treadmill speed.

Real Time Estimation of Energy Expenditure during Physical Activity Using Ensemble Methods

Regular and sufficient amounts of physical activity (PA) are significant in increasing health benefits and mitigating health risks. Given the growing popularity of wrist-worn devices across all age groups, a rigorous evaluation for recognizing hallmark measures of physical activities and estimating energy expenditure is needed to compare their accuracy across the lifespan. The goal of the study was to build machine learning models to recognizing the hallmark measures of PA and estimating energy expenditure (EE), and to test the hypothesis that model performance varies across age-group: young [20-50 years], middle (50-70 years], and old (70-89 years]. Participants (n = 253, 62% women, aged 20-89 years old) performed a battery of 33 daily activities in a standardized laboratory setting while wearing a portable metabolic unit to measure EE that was used to gauge metabolic intensity. Participants also wore a Tri-axial accelerometer on the right wrist. Results from random forests algorithm were quite accurate at recognizing PA type; the F1-Score range across age groups was: sedentary [0.955 – 0.973], locomotion [0.942 – 0.964], and lifestyle [0.913 – 0.949]. Recognizing PA intensity resulted in lower performance; the F1-Score range across age groups was: sedentary [0.919 – 0.947], light [0.813 – 0.828], and moderate [0.846–0.875]. The root mean square error range was [0.835–1.009] for the estimation of EE. The F1-Score range for recognizing individual PAs was [0.263–0.784]. In conclusion, machine learning models used to represent accelerometry data are robust to age differences and a generalizable approach might be sufficient to utilize in accelerometer-based wearables.

ABSTRACTObjective: Accurate estimation of resting energy expenditure (REE) in childrenand adolescents is important to establish estimated energy requirements. The aim of the present study was to measure REE in obese children and adolescents by indirect calorimetry method, compare these values with REE values estimated by equations, and develop the most appropriate equation for this group.Methods: One hundred and three obese children and adolescents (57 males, 46 females) between 7 and 17 years (10.6 ± 2.19 years) were recruited for the study. REE measurements of subjects were made with indirect calorimetry (COSMED, FitMatePro, Rome, Italy) and body compositions were analyzed.Results: In females, the percentage of accurate prediction varied from 32.6 (World Health Organization [WHO]) to 43.5 (Molnar and Lazzer). The bias for equations was −0.2% (Kim), 3.7% (Molnar), and 22.6% (Derumeaux-Burel). Kim's (266 kcal/d), Schmelzle's (267 kcal/d), and Henry's equations (268 kcal/d) had the lowest root mean square error (RMSE; respectively 266, 267, 268 kcal/d). The equation that has the highest RMSE values among female subjects was the Derumeaux-Burel equation (394 kcal/d). In males, when the Institute of Medicine (IOM) had the lowest accurate prediction value (12.3%), the highest values were found using Schmelzle's (42.1%), Henry's (43.9%), and Müller's equations (fat-free mass, FFM; 45.6%). When Kim and Müller had the smallest bias (−0.6%, 9.9%), Schmelzle's equation had the smallest RMSE (331 kcal/d). The new specific equation based on FFM was generated as follows: REE = 451.722 + (23.202 * FFM). According to Bland-Altman plots, it has been found out that the new equations are distributed randomly in both males and females.Conclusion: Previously developed predictive equations mostly provided unaccurate and biased estimates of REE. However, the new predictive equations allow clinicians to estimate REE in an obese children and adolescents with sufficient and acceptable accuracy.

Energy expenditure during daily activities as measured by two motion sensors in patients with COPD

Dear Editor in Chief, We read with interest the recent article by Potter et al. (1) that represents a new opportunity for methodological advances in using global positioning systems (GPS) to estimate outdoor walking energy expenditure (EE). The authors proposed a method that consisted of smoothing GPS-based EE estimates to better match EE estimated by a mobile metabolic sensor in non–steady-state conditions. The theoretical basis for this method was supported by the authors as follows: “The delayed respiratory response to physical activity […] complicates the use of walking equations on a second-by-second basis. Energy expenditure will be overestimated initially before a respiratory steady state is reached.” This assumption is physiologically wrong, and that is probably why it is not supported by any specific reference cited in their article. In fact, the transient change and non–steady-state V˙O2 at the beginning of a given physical activity, or when physical activity intensity varies, means that gas exchange measurements are, by their very nature, limited to estimate actual EE as long as a steady state is not reached (2). Unless considering that the actual EE needed for a given work is different between the beginning of walking and a couple of minutes (3–4 min) after, EE estimated by walking equations should better reflect actual EE than EE estimated from gas exchange during the transitional non–steady-state period. Thus, to our opinion, the right question should be “How to accurately estimate actual EE during conditions of non–steady-state exercise during which EE estimated by gas exchange measurements is flawed?,” and not “How to modify EE estimated by walking equations to better match EE estimated by gas exchange measurements during non-steady state exercise?.” The flawed theoretical basis supported by Potter et al. (1) makes difficult the analysis of their results regarding the comparison of the accuracy of three speed- and grade-based equations (3–6) using GPS for estimating outdoor walking EE on complex terrain. Further, additional limitations prevent unbiased comparisons between the EE estimations achieved by each equation, including (i) the absence of information regarding the precision of the GPS device used and the method for GPS speed computation; (ii) the use of an original equation (5) to estimate outdoor walking EE during downhill walking, whereas such estimations were not allowed by that original equation; and (iii) the conversion of V˙O2 data to watts, which inevitably added error when cross validating the two equations that were initially developed to estimate METs or V˙O2. In fact, there is little doubt that the Pandolf–Santee equation (3,4) would have better fitted actual EE as compared with the two other equations tested, all the more so as this equation includes a specific component for extra load carriage, which was not the case for the two other equations. Pierre-Yves de Müllenheim Bénédicte Noury-Desvaux Institute of Physical Education and Sports Sciences (IFEPSA)West Catholic University (UCO)Les Ponts-de-Cé, FRANCE Laboratory of Exercise Physiology and Vascular MedicineUniversity HospitalAngers, FRANCE Alexis Le Faucheur Univ Rennes, M2S-EA 7470Rennes, FRANCE Clinical Investigation CenterINSERM CIC 1414Rennes, FRANCE