A machine learning approach to measure and monitor physical activity in children to help fight overweight and obesity

Physical Activity is important for maintaining healthy lifestyles. Recommendations for physical activity levels are issued by most governments as part of public health measures. As such, reliable measurement of physical activity for regulatory purposes is vital. This has lead research to explore standards for achieving this using wearable technology and artificial neural networks that produce classifications for specific physical activity events. Applied from a very early age, the ubiquitous capture of physical activity data using mobile and wearable technology may help us to understand how we can combat childhood obesity and the impact that this has in later life. A supervised machine learning approach is adopted in this paper that utilizes data obtained from accelerometer sensors worn by children in free-living environments. The paper presents a set of activities and features suitable for measuring physical activity and evaluates the use of a Multilayer Perceptron neural network to classify physical activities by activity type. A rigorous reproducible data science methodology is presented for subsequent use in physical activity research. Our results show that it was possible to obtain an overall accuracy of 96 % with 95 % for sensitivity, 99 % for specificity and a kappa value of 94 % when three and four feature combinations were used.

An adaptive orientation misalignment calibration method for shoulder movements using inertial sensors : a feasibility study

Qualitative assessment of the progress in physical rehabilitation largely depends on accurate measurement of the range of movements and other kinematic parameters. In clinical practice, wearable inertial sensors have proved to be a potential candidate for such measurements, over the traditional marker based optical systems due to cost and space considerations. The accuracy of wearable sensors have a significant dependence on the initial orientation calibration and the assumption that the sensor will not slip or move with respect to the attached limb. This article introduces a novel calibration algorithm to correct initial orientation misalignment, as well as to track and correct subsequent alignment errors progressively throughout the experiment. The theoretical assertions are validated through controlled experiments with simulated accelerometer and gyroscope measurements.

Interpreting behaviors from accelerometry: a method combining simplicity and objectivity.

Quantifying the behavior of motile, free-ranging animals is difficult. The accelerometry technique offers a method for recording behaviors but interpretation of the data is not straightforward. To date, analysis of such data has either involved subjective, study-specific assignments of behavior to acceleration data or the use of complex analyses based on machine learning. Here, we present a method for automatically classifying acceleration data to represent discrete, coarse-scale behaviors. The method centers on examining the shape of histograms of basic metrics readily derived from acceleration data to objectively determine threshold values by which to separate behaviors. Through application of this method to data collected on two distinct species with greatly differing behavioral repertoires, kittiwakes, and humans, the accuracy of this approach is demonstrated to be very high, comparable to that reported for other automated approaches already published. The method presented offers an alternative to existing methods as it uses biologically grounded arguments to distinguish behaviors, it is objective in determining values by which to separate these behaviors, and it is simple to implement, thus making it potentially widely applicable. The R script coding the method is provided.

Measuring the accuracy of self-reported height and weight in a community-based sample of young people

BACKGROUND: Self-reported anthropometric data are commonly used to estimate prevalence of obesity in population and community-based studies. We aim to: 1) Determine whether survey participants are able and willing to self-report height and weight; 2) Assess the accuracy of self-reported compared to measured anthropometric data in a community-based sample of young people.

METHODS: Participants (16-29 years) of a behaviour survey, recruited at a Melbourne music festival (January 2011), were asked to self-report height and weight; researchers independently weighed and measured a sub-sample. Body Mass Index was calculated and overweight/obesity classified as ≥25 kg/m². Differences between measured and self-reported values were assessed using paired t-test/Wilcoxon signed ranks test. Accurate report of height and weight were defined as <2 cm and <2 kg difference between self-report and measured values, respectively. Agreement between classification of overweight/obesity by self-report and measured values was assessed using McNemar's test.

RESULTS: Of 1405 survey participants, 82% of males and 72% of females self-reported their height and weight. Among 67 participants who were also independently measured, self-reported height and weight were significantly less than measured height (p=0.01) and weight (p<0.01) among females, but no differences were detected among males. Overall, 52% accurately self-reported height, 30% under-reported, and 18% over-reported; 34% accurately self-reported weight, 52% under-reported and 13% over-reported. More females (70%) than males (35%) under-reported weight (p=0.01). Prevalence of overweight/obesity was 33% based on self-report data and 39% based on measured data (p=0.16).

CONCLUSIONS: Self-reported measurements may underestimate weight but accurately identified overweight/obesity in the majority of this sample of young people.

Validation of a body-worn accelerometer to measure activity patterns in octogenarians

OBJECTIVE: To determine the validity of a triaxial body-worn accelerometer for detection of gait and postures in people aged >80 years. DESIGN: Participants performed a range of activities (sitting, lying, walking, standing) in both a controlled and a home setting while wearing the accelerometer. Activities in the controlled setting were performed in a scripted sequence. Activities in the home setting were performed in an unscripted manner. Analyzed accelerometer data were compared against video observation as the reference measure. SETTING: Independent-living and long-term-care retirement village. PARTICIPANTS: Older people (N=22; mean age ± SD, 88.1±5y) residing in long-term-care and independent-living retirement facilities. INTERVENTIONS: Not applicable. MAIN OUTCOME MEASURES: The level of agreement between video observation and the accelerometer for the total duration of each activity, and second-by-second correspondence between video observation and the accelerometer for each activity. RESULTS: The median absolute percentage errors between video observation and the accelerometer were <1% for locomotion and lying. The absolute percentage errors were higher for sitting (median, -22.3%; interquartile range [IQR], -62.8% to 10.7%) and standing (median, 24.7%; IQR, -7.3% to 39.6%). A second-by-second analysis between video observation and the accelerometer found an overall agreement of ≥85% for all activities except standing (median, 56.1%; IQR, 34.8%-81.2%). CONCLUSIONS: This single-device accelerometer provides a valid measure of lying and locomotion in people aged >80 years. There is an error of approximately 25% when discriminating sitting from standing postures, which needs to be taken into account when monitoring longer-term habitual activity in this age group.

Estimating energy expenditure with the RT3 triaxial accelerometer

The RT3 is a relatively new triaxial accelerometer that has replaced the TniTrac. The aim of this study was to validate the RT3 against doubly labeled water (DLW) in a free-living, mixed weight sample of adults. Total energy expenditure (TEE) was measured over a 15-day period using DLW Activity-related energy expenditure (AEE) was estimated by subtracting resting energy expenditure and thermic effect of feeding from TEE. The RT3 triaxial accelerometer was worn over 14 consecutive days. TEE and AEE were estimated using the RT3 proprietary equation. Thirty-six adults ages 18-56 years (56% women) with an average weight of 75.9 kg (SD = 14.8) completed all measurements. Compared to DLW the RT3 underestimated TEE by 539 kJ (4%) and AEE by 485 kJ (15%) on average. The RT3 provided a relatively accurate assessment of free-living activity-related energy expenditure at the group level and generally underestimated total and activity-related energy expenditure compared to DLW

Improving load forecast accuracy by clustering consumers using smart meter data

Utility companies provide electricity to a large number of consumers. These companies need to have an accurate forecast of the next day electricity demand. Any forecast errors will result in either reliability issues or increased costs for the company. Because of the widespread roll-out of smart meters, a large amount of high resolution consumption data is now accessible which was not available in the past. This new data can be used to improve the load forecast and as a result increase the reliability and decrease the expenses of electricity providers. In this paper, a number of methods for improving load forecast using smart meter data are discussed. In these methods, consumers are first divided into a number of clusters. Then a neural network is trained for each cluster and forecasts of these networks are added together in order to form the prediction for the aggregated load. In this paper, it is demonstrated that clustering increases the forecast accuracy significantly. Criteria used for grouping consumers play an important role in this process. In this work, three different feature selection methods for clustering consumers are explained and the effect of feature extraction methods on forecast error is investigated.

Applicability and accuracy of the initially dry and initially wet contact filter paper tests for matric suction measurement of geosynthetic clay liners.

An initially wet contact filter paper test (IW-CFPT) and an initially dry contact filter paper test (ID-CFPT) were used to examine the wetting paths of geosynthetic clay liners, including non-contact filter paper tests for comparative purposes. The CFPTs were applied to both geosynthetic clay liner faces to examine the effect of geotextile type on capillary contact. The non-woven geotextile face was found to be more likely to cause capillary breaks than the woven geotextile face. Both IW- and ID-CFPTs were found to be applicable to geosynthetic clay liners within their accurate upper matric suction measurement limits of 146 kPa and 66 kPa, respectively.

Improvement of accuracy in a free bending test for material characterization

The effect of secondary (anticlastic) curvature and the stress state on the measurement of material properties in a free bending test is studied in order to improve the accuracy of the test. Experiments and numerical analysis are conducted on a medium strength 304L stainless steel and high strength dual-phase steels, DP780 and DP1000. The dependence of the secondary curvature on sample geometry is analysed and correction factors are introduced to improve the accuracy of the calculation of material properties when using plane strain or uniaxial stress two-dimensional assumptions. A free bending test procedure is proposed to characterize material behaviour close to yield. This will allow the quick and simple analysis of material properties for bending-dominated forming processes such as roll forming.

The creation and evaluation of novel canine training aids for cocaine using molecularly encapsulated sol-gel polymers and an investigation of canine field accuracy

Biological detectors, such as canines, are valuable tools used for the rapid identification of illicit materials. However, recent increased scrutiny over the reliability, field accuracy, and the capabilities of each detection canine is currently being evaluated in the legal system. For example, the Supreme Court case, State of Florida v. Harris, discussed the need for continuous monitoring of canine abilities, thresholds, and search capabilities. As a result, the fallibility of canines for detection was brought to light, as well as a need for further research and understanding of canine detection. This study is two-fold, as it looks to not only create new training aids for canines that can be manipulated for dissipation control, but also investigates canine field accuracy to objects with similar odors to illicit materials. ^ It was the goal of this research to improve upon current canine training aid mimics. Sol-gel polymer training aids, imprinted with the active odor of cocaine, were developed. This novel training aid improved upon the longevity of currently existing training aids, while also provided a way to manipulate the polymer network to alter the dissipation rate of the imprinted active odors. The manipulation of the polymer network could allow handlers to control the abundance of odors presented to their canines, familiarizing themselves to their canine’s capabilities and thresholds, thereby increasing the canines’ strength in court.^ The field accuracy of detection canines was recently called into question during the Supreme Court case, State of Florida v. Jardines, where it was argued that if cocaine’s active odor, methyl benzoate, was found to be produced by the popular landscaping flower, snapdragons, canines will false alert to said flowers. Therefore, snapdragon flowers were grown and tested both in the laboratory and in the field to determine the odors produced by snapdragon flowers; the persistence of these odors once flowers have been cut; and whether detection canines will alert to both growing and cut flowers during a blind search scenario. Results revealed that although methyl benzoate is produced by snapdragon flowers, certified narcotics detection canines can distinguish cocaine’s odor profile from that of snapdragon flowers and will not alert.^

Improvement of the prediction accuracy and efficiency of hot strength of austenitic steels with optimised ANN training schemes

The hot strength of austenitic steels of different carbon contents was modelled using an artificial neural network (ANN) model with optimum training data. As training data employed in a traditional neural network model were randomly selected from experimental data, they were not representative and the prediction accuracy and efficiency were therefore significantly affected. In this work, only representatively experimental data were used for training and during the procedure, one tenth of the training data extracted from experiment were used for testing the training model and terminating the modelling. The effects of the carbon con tent on flow stress, peak strains and peak stresses observed from the experiment for both training and test data were accurately represented with the ANN scheme reported in this work.

Criterion validity of the activPAL™ and ActiGraph for assessing children's sitting and standing time in a school classroom setting

BACKGROUND: Few studies have investigated the accuracy of the ActiGraph (AG) GTX3 accelerometer for assessing children's sitting and standing time. The activPAL (aP) has an inclinometer function that enables it to distinguish between sitting/lying and standing; however, its accuracy for assessing sitting and standing in older children is unknown. This study validated the accuracy of these devices for estimating sitting and standing time in a school classroom against a criterion measure of direct observation (DO).

FINDINGS: Forty children in grades 5-7 wore both devices while being video recorded during two school lessons. AG and aP data were simultaneously collected in 15-s epochs. Individual participant DO and aP data were recorded as total time spent sitting/lying, standing and stepping. AG data were converted into time spent sitting and standing using previously established cut-points. Compared with DO, the aP underestimated sitting time (mean bias = -1.9 min, 95 % LoA = -8.9 to 5.2 min) and overestimated standing time (mean bias = 1.8 min, 95% LoA = -9.6 to 13.3 min). The best-performing AG cut-point across both sitting and standing (<75 counts/15 s) was more accurate than the aP, underestimating sitting time (mean bias = -0.8 min, 95 % LoA = -10.5 to 9.9 min) and standing time (mean bias = -0.4 min, 95% LoA = -9.8 to 9.1 min), but was less precise as evidenced by wider LoAs and poorer correlations with DO (sitting r = 0.86 aP vs 0.80 AG; standing r = 0.78 aP vs 0.60 AG).

CONCLUSIONS: The aP demonstrated good accuracy and precision for assessing free-living sitting and standing time in classroom settings. The AG was most accurate using a cut-point of < 75 counts/15 s. Further studies should validate the monitors in settings with greater inter- and intra-individual variation in movement patterns.

Accuracy and usefulness of select methods for assessing complete collection of 24-hour urine: a systematic review.

Twenty-four-hour urine collection is the recommended method for estimating sodium intake. To investigate the strengths and limitations of methods used to assess completion of 24-hour urine collection, the authors systematically reviewed the literature on the accuracy and usefulness of methods vs para-aminobenzoic acid (PABA) recovery (referent). The percentage of incomplete collections, based on PABA, was 6% to 47% (n=8 studies). The sensitivity and specificity for identifying incomplete collection using creatinine criteria (n=4 studies) was 6% to 63% and 57% to 99.7%, respectively. The most sensitive method for removing incomplete collections was a creatinine index <0.7. In pooled analysis (≥2 studies), mean urine creatinine excretion and volume were higher among participants with complete collection (P<.05); whereas, self-reported collection time did not differ by completion status. Compared with participants with incomplete collection, mean 24-hour sodium excretion was 19.6 mmol higher (n=1781 specimens, 5 studies) in patients with complete collection. Sodium excretion may be underestimated by inclusion of incomplete 24-hour urine collections. None of the current approaches reliably assess completion of 24-hour urine collection.

Validación de criterio concurrente del sistema de observación de tiempo de instrucción de condición física – SOFIT- para la medición de la actividad física durante las clases de educación física en escolares de 1º A 9º de Bogotá, Colombia

Determinar la validez concurrente del Sistema de Observación de Tiempo de Instrucción de Condición Física (SOFIT) a través de acelerometría, como método para medir los niveles de actividad física (AF) de los escolares de 1º a 9º durante la clase de educación física en tres colegios públicos de Bogotá, Colombia. Estudio transversal entre Octubre de 2014 y Marzo de 2015. La medición se realizó en tres colegios públicos de Bogotá. Participaron 48 estudiantes (25 niñas; 23 niños), entre 5 y 17 años, seleccionados de acuerdo al protocolo de SOFIT. El resultado se categoriza en porcentaje de tiempo en comportamiento sedentario, AF moderada, AF vigorosa, y AF moderada a vigorosa. Se validó utilizando como patrón de oro la acelerometría en las mismas categorías. Se realizó diferencia de medias, regresión lineal y modelo de efectos fijos. La correlación entre SOFIT y acelerometría fue buena para AF moderada (rho=,958; p=0,000), AF vigorosa (rho=,937; p=0,000) y AF de moderada a vigorosa (rho=0,962; p=0,000). Al igual que utilizando un modelo de efectos fijos, AF moderada (β1=0,92; p=0,00), vigorosa (β1=0,94; p=0,00) y AF de moderada a vigorosa (β1=0,95; p=0,00), mostrando ausencia de diferencias significativas entre los dos métodos para la medición de los niveles de AF. El comportamiento sedentario correlacionó positivamente en Spearman (rho=,0965; p=0,000), El sistema SOFIT demostró ser válido para medir niveles de AF en clases de educación física, tras buena correlación y concordancia con acelerometría. SOFIT es un instrumento de fácil acceso y de bajo costo para la medición de la AF durante las clases de educación física en el contexto escolar y se recomienda su uso en futuros estudios.