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

Abstract Background There is considerable promise for using commercial wearable devices for measuring physical activity at the population level. The objective of this study was to examine whether commercial wearable devices could accurately predict lying, sitting, and intensity level of other activities in a lab-based protocol. Methods We recruited a convenience sample of 49 participants (23 men and 26 women) to wear three devices, an Apple Watch Series 2, a Fitbit Charge HR2, and and iPhone 6S. Participants completed a 65-minute protocol consisting of 40 minutes of total treadmill time and 25 minutes of sitting or lying time. Indirect calorimetry was used to measure energy expenditure. The outcome variable for the study was the activity class; lying, sitting, walking self-paced, and running 3 METs, 5 METs, and 7 METs. Minute-by-minute heart rate, steps, distance, and calories from Apple Watch and Fitbit were included in four different machine learning models. Results Our dataset included 3656 and 2608 minutes of Apple Watch and Fitbit data, respectively. We tested decision trees, support vector machines, random forest, and rotation forest models. Rotation forest models had the highest classification accuracies at 82.6% for Apple Watch and 89.3% for Fitbit. Classification accuracies for Apple Watch data ranged from 72.5% for sitting to 89.0% for 7 METs . For Fitbit, accuracies varied between 86.2% for sitting to 92.6% for 7 METs . Conclusion This study demonstrated that commercial wearable devices, Apple Watch and Fitbit, were able to predict physical activity types with a reasonable accuracy. The results support the use of minute-by-minute data from Apple Watch and Fitbit combined with machine learning approaches for scalable physical activity type classification at the population level.

Download Full-text

Predicting lying, sitting, walking and running using Apple Watch and Fitbit data

BMJ Open Sport & Exercise Medicine ◽

10.1136/bmjsem-2020-001004 ◽

2021 ◽

Vol 7 (1) ◽

pp. e001004

Author(s):

Daniel Fuller ◽

Javad Rahimipour Anaraki ◽

Bongai Simango ◽

Machel Rayner ◽

Faramarz Dorani ◽

...

Keyword(s):

Random Forest ◽

Population Level ◽

Wearable Devices ◽

Support Vector ◽

Convenience Sample ◽

Rotation Forest ◽

Metabolic Equivalents ◽

Forest Models ◽

Random Forest Models ◽

Movement Type

ObjectivesThis study’s objective was to examine whether commercial wearable devices could accurately predict lying, sitting and varying intensities of walking and running.MethodsWe recruited a convenience sample of 49 participants (23 men and 26 women) to wear three devices, an Apple Watch Series 2, a Fitbit Charge HR2 and iPhone 6S. Participants completed a 65 min protocol consisting of 40 min of total treadmill time and 25 min of sitting or lying time. The study’s outcome variables were six movement types: lying, sitting, walking self-paced and walking/running at 3 metabolic equivalents of task (METs), 5 METs and 7 METs. All analyses were conducted at the minute level with heart rate, steps, distance and calories from Apple Watch and Fitbit. These included three different machine learning models: support vector machines, Random Forest and Rotation forest.ResultsOur dataset included 3656 and 2608 min of Apple Watch and Fitbit data, respectively. Rotation Forest models had the highest classification accuracies for Apple Watch at 82.6%, and Random Forest models had the highest accuracy for Fitbit at 90.8%. Classification accuracies for Apple Watch data ranged from 72.6% for sitting to 89.0% for 7 METs. For Fitbit, accuracies varied between 86.2% for sitting to 92.6% for 7 METs.ConclusionThis preliminary study demonstrated that data from commercial wearable devices could predict movement types with reasonable accuracy. More research is needed, but these methods are a proof of concept for movement type classification at the population level using commercial wearable device data.

Download Full-text

Phenotyping Women Based on Dietary Macronutrients, Physical Activity, and Body Weight Using Machine Learning Tools

Nutrients ◽

10.3390/nu11071681 ◽

2019 ◽

Vol 11 (7) ◽

pp. 1681 ◽

Cited By ~ 1

Author(s):

Ramyaa Ramyaa ◽

Omid Hosseini ◽

Giri P. Krishnan ◽

Sridevi Krishnan

Keyword(s):

Physical Activity ◽

Machine Learning ◽

Cluster Analysis ◽

Body Weight ◽

Energy Balance ◽

Numerical Data ◽

Support Vector ◽

Learning Tools ◽

K Nearest Neighbor ◽

Numerical Predictions

Nutritional phenotyping can help achieve personalized nutrition, and machine learning tools may offer novel means to achieve phenotyping. The primary aim of this study was to use energy balance components, namely input (dietary energy intake and macronutrient composition) and output (physical activity) to predict energy stores (body weight) as a way to evaluate their ability to identify potential phenotypes based on these parameters. From the Women’s Health Initiative Observational Study (WHI OS), carbohydrates, proteins, fats, fibers, sugars, and physical activity variables, namely energy expended from mild, moderate, and vigorous intensity activity, were used to predict current body weight (both as body weight in kilograms and as a body mass index (BMI) category). Several machine learning tools were used for this prediction. Finally, cluster analysis was used to identify putative phenotypes. For the numerical predictions, the support vector machine (SVM), neural network, and k-nearest neighbor (kNN) algorithms performed modestly, with mean approximate errors (MAEs) of 6.70 kg, 6.98 kg, and 6.90 kg, respectively. For categorical prediction, SVM performed the best (54.5% accuracy), followed closely by the bagged tree ensemble and kNN algorithms. K-means cluster analysis improved prediction using numerical data, identified 10 clusters suggestive of phenotypes, with a minimum MAE of ~1.1 kg. A classifier was used to phenotype subjects into the identified clusters, with MAEs <5 kg for 15% of the test set (n = ~2000). This study highlights the challenges, limitations, and successes in using machine learning tools on self-reported data to identify determinants of energy balance.

Download Full-text

A Machine Learning Classifier for Detection of Physical Activity Types and Postures During Free-Living

Journal for the Measurement of Physical Behaviour ◽

10.1123/jmpb.2021-0015 ◽

2021 ◽

pp. 1-8

Author(s):

Kerstin Bach ◽

Atle Kongsvold ◽

Hilde Bårdstu ◽

Ellen Marie Bardal ◽

Håkon S. Kjærnli ◽

...

Keyword(s):

Physical Activity ◽

Machine Learning ◽

Ground Truth ◽

Daily Physical Activity ◽

Gradient Boosting ◽

Free Living ◽

Recording Time ◽

Learning Classifier ◽

Extreme Gradient Boosting ◽

Activity Types

Introduction: Accelerometer-based measurements of physical activity types are commonly used to replace self-reports. To advance the field, it is desirable that such measurements allow accurate detection of key daily physical activity types. This study aimed to evaluate the performance of a machine learning classifier for detecting sitting, standing, lying, walking, running, and cycling based on a dual versus single accelerometer setups during free-living. Methods: Twenty-two adults (mean age [SD, range] 38.7 [14.4, 25–68] years) were wearing two Axivity AX3 accelerometers positioned on the low back and thigh along with a GoPro camera positioned on the chest to record lower body movements during free-living. The labeled videos were used as ground truth for training an eXtreme Gradient Boosting classifier using window lengths of 1, 3, and 5 s. Performance of the classifier was evaluated using leave-one-out cross-validation. Results: Total recording time was ∼38 hr. Based on 5-s windowing, the overall accuracy was 96% for the dual accelerometer setup and 93% and 84% for the single thigh and back accelerometer setups, respectively. The decreased accuracy for the single accelerometer setup was due to a poor precision in detecting lying based on the thigh accelerometer recording (77%) and standing based on the back accelerometer recording (64%). Conclusion: Key daily physical activity types can be accurately detected during free-living based on dual accelerometer recording, using an eXtreme Gradient Boosting classifier. The overall accuracy decreases marginally when predictions are based on single thigh accelerometer recording, but detection of lying is poor.

Download Full-text

Comparison of the Validity and Generalizability of Machine Learning Algorithms for the Prediction of Energy Expenditure: Validation Study

JMIR mhealth and uhealth ◽

10.2196/23938 ◽

2021 ◽

Vol 9 (8) ◽

pp. e23938

Author(s):

Ruairi O'Driscoll ◽

Jake Turicchi ◽

Mark Hopkins ◽

Cristiana Duarte ◽

Graham W Horgan ◽

...

Keyword(s):

Physical Activity ◽

Machine Learning ◽

Neural Networks ◽

Random Forest ◽

Energy Expenditure ◽

Superior Performance ◽

Gradient Boosting ◽

Support Vector ◽

Data Sets ◽

Out Of Sample

Background Accurate solutions for the estimation of physical activity and energy expenditure at scale are needed for a range of medical and health research fields. Machine learning techniques show promise in research-grade accelerometers, and some evidence indicates that these techniques can be applied to more scalable commercial devices. Objective This study aims to test the validity and out-of-sample generalizability of algorithms for the prediction of energy expenditure in several wearables (ie, Fitbit Charge 2, ActiGraph GT3-x, SenseWear Armband Mini, and Polar H7) using two laboratory data sets comprising different activities. Methods Two laboratory studies (study 1: n=59, age 44.4 years, weight 75.7 kg; study 2: n=30, age=31.9 years, weight=70.6 kg), in which adult participants performed a sequential lab-based activity protocol consisting of resting, household, ambulatory, and nonambulatory tasks, were combined in this study. In both studies, accelerometer and physiological data were collected from the wearables alongside energy expenditure using indirect calorimetry. Three regression algorithms were used to predict metabolic equivalents (METs; ie, random forest, gradient boosting, and neural networks), and five classification algorithms (ie, k-nearest neighbor, support vector machine, random forest, gradient boosting, and neural networks) were used for physical activity intensity classification as sedentary, light, or moderate to vigorous. Algorithms were evaluated using leave-one-subject-out cross-validations and out-of-sample validations. Results The root mean square error (RMSE) was lowest for gradient boosting applied to SenseWear and Polar H7 data (0.91 METs), and in the classification task, gradient boost applied to SenseWear and Polar H7 was the most accurate (85.5%). Fitbit models achieved an RMSE of 1.36 METs and 78.2% accuracy for classification. Errors tended to increase in out-of-sample validations with the SenseWear neural network achieving RMSE values of 1.22 METs in the regression tasks and the SenseWear gradient boost and random forest achieving an accuracy of 80% in classification tasks. Conclusions Algorithms trained on combined data sets demonstrated high predictive accuracy, with a tendency for superior performance of random forests and gradient boosting for most but not all wearable devices. Predictions were poorer in the between-study validations, which creates uncertainty regarding the generalizability of the tested algorithms.

Download Full-text

A Supervised Machine Learning Approach to Detect the On/Off State in Parkinson’s Disease Using Wearable Based Gait Signals

Diagnostics ◽

10.3390/diagnostics10060421 ◽

2020 ◽

Vol 10 (6) ◽

pp. 421

Author(s):

Satyabrata Aich ◽

Jinyoung Youn ◽

Sabyasachi Chakraborty ◽

Pyari Mohan Pradhan ◽

Jin-han Park ◽

...

Keyword(s):

Machine Learning ◽

Parkinson’S Disease ◽

Parkinson's Disease ◽

Random Forest ◽

Wearable Devices ◽

Supervised Machine Learning ◽

Machine Learning Techniques ◽

Support Vector ◽

Healthcare Applications ◽

Reported Data

Fluctuations in motor symptoms are mostly observed in Parkinson’s disease (PD) patients. This characteristic is inevitable, and can affect the quality of life of the patients. However, it is difficult to collect precise data on the fluctuation characteristics using self-reported data from PD patients. Therefore, it is necessary to develop a suitable technology that can detect the medication state, also termed the “On”/“Off” state, automatically using wearable devices; at the same time, this could be used in the home environment. Recently, wearable devices, in combination with powerful machine learning techniques, have shown the potential to be effectively used in critical healthcare applications. In this study, an algorithm is proposed that can detect the medication state automatically using wearable gait signals. A combination of features that include statistical features and spatiotemporal gait features are used as inputs to four different classifiers such as random forest, support vector machine, K nearest neighbour, and Naïve Bayes. In total, 20 PD subjects with definite motor fluctuations have been evaluated by comparing the performance of the proposed algorithm in association with the four aforementioned classifiers. It was found that random forest outperformed the other classifiers with an accuracy of 96.72%, a recall of 97.35%, and a precision of 96.92%.

Download Full-text

Comparison of Machine Learning Algorithms for the Quality Assessment of Wearable ECG Signals Via Lenovo H3 Devices

Journal of Medical and Biological Engineering ◽

10.1007/s40846-020-00588-7 ◽

2021 ◽

Author(s):

Fan Fu ◽

Wentao Xiang ◽

Yukun An ◽

Bin Liu ◽

Xianqing Chen ◽

...

Keyword(s):

Machine Learning ◽

Heart Disease ◽

Quality Assessment ◽

Learning Algorithms ◽

Wearable Devices ◽

Machine Learning Algorithms ◽

Assessment System ◽

Support Vector ◽

Signal Quality ◽

Ecg Signals

Abstract Purpose Electrocardiogram (ECG) signals collected from wearable devices are easily corrupted with surrounding noise and artefacts, where the signal-to-noise ratio (SNR) of wearable ECG signals is significantly lower than that from hospital ECG machines. To meet the requirements for monitoring heart disease via wearable devices, eliminating useless or poor-quality ECG signals (e.g., lead-falls and low SNRs) can be solved by signal quality assessment algorithms. Methods To compensate for the deficiency of the existing ECG quality assessment system, a wearable ECG signal dataset from heart disease patients collected by Lenovo H3 devices was constructed. Then, this paper compares the performance of three machine learning algorithms, i.e., the traditional support vector machine (SVM), least-squares SVM (LS-SVM) and long short-term memory (LSTM) algorithms. Different non-morphological signal quality indices (i.e., the approximate entropy (ApEn), sample entropy (SaEn), fuzzy measure entropy (FMEn), Hurst exponent (HE), kurtosis (K) and power spectral density (PSD) features) extracted from the original ECG signals are fed into the three algorithms as input. Results The true positive rate, true negative rate, sensitivity and accuracy are used to evaluate the performance of each method, and the LSTM algorithm achieves the best results on these metrics (97.14%, 86.8%, 97.46% and 95.47%, respectively). Conclusions Among the three algorithms, the LSTM-based quality assessment method is the most suitable for the signals collected by the Lenovo H3 devices. The results also show that the combination of statistical features can effectively evaluate the quality of ECG signals.

Download Full-text

Clusters of Adolescent Physical Activity Tracker Patterns and Their Associations With Physical Activity Behaviors in Finland and Ireland: Cross-Sectional Study

Journal of Medical Internet Research ◽

10.2196/18509 ◽

2020 ◽

Vol 22 (9) ◽

pp. e18509 ◽

Cited By ~ 1

Author(s):

Kwok Ng ◽

Sami Kokko ◽

Tuija Tammelin ◽

Jouni Kallio ◽

Sarahjane Belton ◽

...

Keyword(s):

Physical Activity ◽

Sport Participation ◽

Vigorous Physical Activity ◽

Population Level ◽

Wearable Devices ◽

Active Travel ◽

Wearable Device ◽

Cross Sectional ◽

Sports Clubs ◽

Males And Females

Background Physical activity trackers (PATs) such as apps and wearable devices (eg, sports watches, heart rate monitors) are increasingly being used by young adolescents. Despite the potential of PATs to help monitor and improve moderate-to-vigorous physical activity (MVPA) behaviors, there is a lack of research that confirms an association between PAT ownership or use and physical activity behaviors at the population level. Objective The purpose of this study was to examine the ownership and use of PATs in youth and their associations with physical activity behaviors, including daily MVPA, sports club membership, and active travel, in 2 nationally representative samples of young adolescent males and females in Finland and Ireland. Methods Comparable data were gathered in the 2018 Finnish School-aged Physical Activity (F-SPA 2018, n=3311) and the 2018 Irish Children’s Sport Participation and Physical Activity (CSPPA 2018, n=4797) studies. A cluster analysis was performed to obtain the patterns of PAT ownership and usage by adolescents (age, 11-15 years). Four similar clusters were identified across Finnish and Irish adolescents: (1) no PATs, (2) PAT owners, (3) app users, and (4) wearable device users. Adjusted binary logistic regression analyses were used to evaluate how PAT clusters were associated with physical activity behaviors, including daily MVPA, membership of sports clubs, and active travel, after stratification by gender. Results The proportion of app ownership among Finnish adolescents (2038/3311, 61.6%) was almost double that of their Irish counterparts (1738/4797, 36.2%). Despite these differences, the clustering patterns of PATs were similar between the 2 countries. App users were more likely to take part in daily MVPA (males, odds ratio [OR] 1.27, 95% CI 1.04-1.55; females, OR 1.49, 95% CI 1.20-1.85) and be members of sports clubs (males, OR 1.37, 95% CI 1.15-1.62; females, OR 1.25, 95% CI 1.07-1.50) compared to the no PATs cluster, after adjusting for country, age, family affluence, and disabilities. These associations, after the same adjustments, were even stronger for wearable device users to participate in daily MVPA (males, OR 1.83, 95% CI 1.49-2.23; females, OR 2.25, 95% CI 1.80-2.82) and be members of sports clubs (males, OR 1.88, 95% CI 1.55-2.88; females, OR 2.07, 95% CI 1.71-2.52). Significant associations were observed between male users of wearable devices and taking part in active travel behavior (OR 1.39, 95% CI 1.04-1.86). Conclusions Although Finnish adolescents report more ownership of PATs than Irish adolescents, the patterns of use and ownership remain similar among the cohorts. The findings of our study show that physical activity behaviors were positively associated with wearable device users and app users. These findings were similar between males and females. Given the cross-sectional nature of this data, the relationship between using apps or wearable devices and enhancing physical activity behaviors requires further investigation.

Download Full-text

A machine learning based exploration of COVID-19 mortality risk

PLoS ONE ◽

10.1371/journal.pone.0252384 ◽

2021 ◽

Vol 16 (7) ◽

pp. e0252384

Author(s):

Mahdi Mahdavi ◽

Hadi Choubdar ◽

Erfan Zabeh ◽

Michael Rieder ◽

Safieddin Safavi-Naeini ◽

...

Keyword(s):

Machine Learning ◽

Demographic Data ◽

Wearable Devices ◽

Mortality Prediction ◽

Support Vector ◽

Novel Technologies ◽

Mortality Risks ◽

Non Invasive ◽

Information Contents ◽

Efficient Resource

Early prediction of patient mortality risks during a pandemic can decrease mortality by assuring efficient resource allocation and treatment planning. This study aimed to develop and compare prognosis prediction machine learning models based on invasive laboratory and noninvasive clinical and demographic data from patients’ day of admission. Three Support Vector Machine (SVM) models were developed and compared using invasive, non-invasive, and both groups. The results suggested that non-invasive features could provide mortality predictions that are similar to the invasive and roughly on par with the joint model. Feature inspection results from SVM-RFE and sparsity analysis displayed that, compared with the invasive model, the non-invasive model can provide better performances with a fewer number of features, pointing to the presence of high predictive information contents in several non-invasive features, including SPO2, age, and cardiovascular disorders. Furthermore, while the invasive model was able to provide better mortality predictions for the imminent future, non-invasive features displayed better performance for more distant expiration intervals. Early mortality prediction using non-invasive models can give us insights as to where and with whom to intervene. Combined with novel technologies, such as wireless wearable devices, these models can create powerful frameworks for various medical assignments and patient triage.

Download Full-text

Stress Classification of ECG-Derived HRV Features Extracted from Wearable Devices

10.20944/preprints202103.0644.v1 ◽

2021 ◽

Author(s):

Kayisan Mary Dalmeida ◽

Giovanni Luca Masala

Keyword(s):

Machine Learning ◽

Nearest Neighbor ◽

Wearable Devices ◽

Mental Wellbeing ◽

Gradient Boosting ◽

Support Vector ◽

K Nearest Neighbor ◽

Automobile Crashes ◽

Machine Learning Model ◽

Stress Classification

Stress has been identified as one of the major causes of automobile crashes which then lead to high rates of fatalities and injuries each year. Stress can be measured via physiological measurements and in this study the focus will be based on the features that can be extracted by common wearable devices. Hence the study will be mainly focusing on the heart rate variability (HRV). This study is aimed to develop a good predictive model that can accurately classify stress levels from ECG-derived HRV features, obtained from automobile drivers, testing different machine learning methodologies such as K-Nearest Neighbor (KNN), Support Vector Machines (SVM), Multilayer Perceptron (MLP), Random Forest (RF) and Gradient Boosting (GB). Moreover, the models obtained with highest predictive power will be used as reference for the development of a machine learning model that would be used to classify stress from HRV features derived from HRV measurements obtained from wearable devices. We demonstrate that MLP was the ideal stress classifier by achieving a Recall of 80%. The proposed method can be also used on all applications in which is important to monitor the stress level e. g. in physical rehabilitation, anxiety relief or mental wellbeing.

Download Full-text

Application of Raw Accelerometer Data and Machine-Learning Techniques to Characterize Human Movement Behavior: A Systematic Scoping Review

Journal of Physical Activity and Health ◽

10.1123/jpah.2019-0088 ◽

2020 ◽

Vol 17 (3) ◽

pp. 360-383 ◽

Cited By ~ 4

Author(s):

Anantha Narayanan ◽

Farzanah Desai ◽

Tom Stewart ◽

Scott Duncan ◽

Lisa Mackay

Keyword(s):

Physical Activity ◽

Machine Learning ◽

Scoping Review ◽

Machine Learning Algorithms ◽

Supervised Machine Learning ◽

Machine Learning Techniques ◽

Movement Behavior ◽

Support Vector ◽

Accelerometer Data ◽

Learning Techniques

Background: Application of machine learning for classifying human behavior is increasingly common as access to raw accelerometer data improves. The aims of this scoping review are (1) to examine if machine-learning techniques can accurately identify human activity behaviors from raw accelerometer data and (2) to summarize the practical implications of these machine-learning techniques for future work. Methods: Keyword searches were performed in Scopus, Web of Science, and EBSCO databases in 2018. Studies that applied supervised machine-learning techniques to raw accelerometer data and estimated components of physical activity were included. Information on study characteristics, machine-learning techniques, and key study findings were extracted from included studies. Results: Of the 53 studies included in the review, 75% were published in the last 5 years. Most studies predicted postures and activity type, rather than intensity, and were conducted in controlled environments using 1 or 2 devices. The most common models were support vector machine, random forest, and artificial neural network. Overall, classification accuracy ranged from 62% to 99.8%, although nearly 80% of studies achieved an overall accuracy above 85%. Conclusions: Machine-learning algorithms demonstrate good accuracy when predicting physical activity components; however, their application to free-living settings is currently uncertain.

Download Full-text