Activity Type Recognition Research Articles

Validation of an Activity Type Recognition Model Classifying Daily Physical Behavior in Older Adults: The HAR70+ Model

Activity monitoring combined with machine learning (ML) methods can contribute to detailed knowledge about daily physical behavior in older adults. The current study (1) evaluated the performance of an existing activity type recognition ML model (HARTH), based on data from healthy young adults, for classifying daily physical behavior in fit-to-frail older adults, (2) compared the performance with a ML model (HAR70+) that included training data from older adults, and (3) evaluated the ML models on older adults with and without walking aids. Eighteen older adults aged 70-95 years who ranged widely in physical function, including usage of walking aids, were equipped with a chest-mounted camera and two accelerometers during a semi-structured free-living protocol. Labeled accelerometer data from video analysis was used as ground truth for the classification of walking, standing, sitting, and lying identified by the ML models. Overall accuracy was high for both the HARTH model (91%) and the HAR70+ model (94%). The performance was lower for those using walking aids in both models, however, the overall accuracy improved from 87% to 93% in the HAR70+ model. The validated HAR70+ model contributes to more accurate classification of daily physical behavior in older adults that is essential for future research.

Machine Learning Models for Weight-Bearing Activity Type Recognition Based on Accelerometry in Postmenopausal Women

Hip-worn triaxial accelerometers are widely used to assess physical activity in terms of energy expenditure. Methods for classification in terms of different types of activity of relevance to the skeleton in populations at risk of osteoporosis are not currently available. This publication aims to assess the accuracy of four machine learning models on binary (standing and walking) and tertiary (standing, walking, and jogging) classification tasks in postmenopausal women. Eighty women performed a shuttle test on an indoor track, of which thirty performed the same test on an indoor treadmill. The raw accelerometer data were pre-processed, converted into eighteen different features and then combined into nine unique feature sets. The four machine learning models were evaluated using three different validation methods. Using the leave-one-out validation method, the highest average accuracy for the binary classification model, 99.61%, was produced by a k-NN Manhattan classifier using a basic statistical feature set. For the tertiary classification model, the highest average accuracy, 94.04%, was produced by a k-NN Manhattan classifier using a feature set that included all 18 features. The methods and classifiers within this study can be applied to accelerometer data to more accurately characterize weight-bearing activity which are important to skeletal health.

Are Machine Learning Models on Wrist Accelerometry Robust against Differences in Physical Performance among Older Adults?

Sufficient physical activity (PA) reduces the risk of a myriad of diseases and preserves physical capabilities in later life. While there have been significant achievements in mapping accelerations to real-life movements using machine learning (ML), errors continue to be common, particularly for wrist-worn devices. It remains unknown whether ML models are robust for estimating age-related loss of physical function. In this study, we evaluated the performance of ML models (XGBoost and LASSO) to estimate the hallmark measures of PA in low physical performance (LPP) and high physical performance (HPP) groups. Our models were built to recognize PA types and intensities, identify each individual activity, and estimate energy expenditure (EE) using wrist-worn accelerometer data (33 activities per participant) from a large sample of participants (n = 247, 57% females, aged 60+ years). Results indicated that the ML models were accurate in recognizing PA by type and intensity while also estimating EE accurately. However, the models built to recognize individual activities were less robust. Across all tasks, XGBoost outperformed LASSO. XGBoost obtained F1-Scores for sedentary (0.932 0.005), locomotion (0.946 0.003), lifestyle (0.927 0.006), and strength flexibility exercise (0.915 0.017) activity type recognition tasks. The F1-Scores for recognizing low, light, and moderate activity intensity were (0.9320.005), (0.840 0.004), and (0.869 0.005), respectively. The root mean square error for EE estimation was 0.836 0.059 METs. There was no evidence showing that splitting the participants into the LPP and HPP groups improved the models’ performance on estimating the hallmark measures of physical activities. In conclusion, using features derived from wrist-worn accelerometer data, machine learning models can accurately recognize PA types and intensities and estimate EE for older adults with high and low physical function.

A Multipurpose Wearable Sensor-Based System for Weight Training

In recent years, there has been growing interest in automated tracking and detection of sports activities. Researchers have shown that providing activity information to individuals during their exercise routines can greatly help them in achieving their exercise goals. In particular, such information would help them to maximize workout efficiency and prevent overreaching and overtraining. This paper presents the development of a novel multipurpose wearable device for automatic weight detection, activity type recognition, and count repetition in sports activities such as weight training. The device monitors weights and activities by using an inertial measurement unit (IMU), an accelerometer, and three force sensors mounted in a glove, and classifies them by utilizing developed machine learning models. For weight detection purposes, different classifiers including Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), and Multi-layer Perceptron Neural Networks (MLP) were investigated. For activity recognition, the K nearest neighbor (KNN), Decision Tree (DT), Random Forest (RF), and SVM models were trained and examined. Experimental results indicate that the SVM classifier can achieve the highest accuracy for weight detection whereas RF can outperform other classifiers for activity recognition. The results indicate feasibility of developing a wearable device that can provide in-situ accurate information regarding the lifted weight and activity type with minimum physical intervention.

Spatiotemporal Information Fusion Method of User and Social Media Activity

Social media check-in data contains a lot of user activity information. Understanding the types of activities and behavior of social media users has important research significance for exploring human mobility and behavior patterns. This paper studies the user activity classification method for Sina Weibo (a very popular Chinese social network service, referred to as “Weibo”), which combines image expression and spatiotemporal data classification technology to realize the identification of the activity behavior represented by the microblog check-in data. Firstly, the user activities represented by the Sina Weibo check-in data are divided into six categories according to POI attribute information: “catering”, “life services”, “campus”, “outdoors”, “entertainment” and “travel”; Then, through the Convolutional Neural Network (CNN) and K-Nearest Neighbor (KNN) classification methods, the image scene information and spatiotemporal information in the check-in data are fused to classify the activity behavior of microblog users. The experimental results show that the proposed method can significantly improve the accuracy of microblog user activity type recognition and provide more effective data support for accurately exploring human behavior activities.

Accelerometer-based physical activity: total volume per day and standardized measures.

The use of accelerometers in physical activity (PA) research has increased exponentially over the past 20 yr. The first commercially available accelerometer for assessing PA, the Caltrac, was worn on the waist and estimated PA energy expenditure in kilocalories. Around 1995, the emphasis shifted to measuring minutes of moderate-to-vigorous PA (MVPA), especially for bouts of 10 min or longer. Recent studies, however, show that light-intensity PA and intermittent (nonbout) MVPA also have important health benefits. The total volume of PA performed is an important variable because it takes the frequency, intensity, and duration of activity bouts and condenses them down into a single metric. The total volume of PA is appropriate for many research applications and can enhance comparisons between studies. In the future, machine learning algorithms will provide improved accuracy for activity type recognition and estimation of PA energy expenditure. However, in the current landscape of objectively measured PA, total activity counts per day (TAC/d) is a proxy for the total volume of PA. TAC/d percentiles for age- and gender-specific groups have been developed from the National Health and Nutrition Examination Survey ActiGraph data (2003-2006), providing a novel way to assess PA. The use of TAC/d or standardized units of acceleration could harmonize PA across studies. TAC/d should be viewed as an additional metric, not intended to replace other metrics (e.g., sedentary time, light-intensity PA, moderate PA, and vigorous PA) that may also be related to health. As future refinements to wearable monitors occur, researchers should continue to consider metrics that reflect the total volume of PA in addition to existing PA metrics.

Validation of five minimally obstructive methods to estimate physical activity energy expenditure in young adults in semi-standardized settings.

We compared the accuracy of five objective methods, including two newly developed methods combining accelerometry and activity type recognition (Acti4), against indirect calorimetry, to estimate total energy expenditure (EE) of different activities in semi-standardized settings. Fourteen participants performed a standardized and semi-standardized protocol including seven daily life activity types, while having their EE measured by indirect calorimetry. Simultaneously, physical activity was quantified by an ActivPAL3, two ActiGraph GT3X+’s and an Actiheart. EE was estimated by the standard ActivPAL3 software (ActivPAL), ActiGraph GT3X+ (ActiGraph) and Actiheart (Actiheart), and by a combination of activity type recognition via Acti4 software and activity counts per minute (CPM) of either a hip- or thigh-worn ActiGraph GT3X+ (AGhip + Acti4 and AGthigh + Acti4). At group level, estimated physical activities EE by Actiheart (MSE = 2.05) and AGthigh + Acti4 (MSE = 0.25) were not significantly different from measured EE by indirect calorimetry, while significantly underestimated by ActiGraph, ActivPAL and AGhip + Acti4. AGthigh + Acti4 and Actiheart explained 77% and 45%, of the individual variations in measured physical activity EE by indirect calorimetry, respectively. This study concludes that combining accelerometer data from a thigh-worn ActiGraph GT3X+ with activity type recognition improved the accuracy of activity specific EE estimation against indirect calorimetry in semi-standardized settings compared to previously validated methods using CPM only.

Extracting Activity-travel Diaries from GPS Data: Towards Integrated Semi-automatic Imputation

This paper presents an integrated approach to extracting activity-travel diaries from GPS data. The imputation involves a semi-automatic procedure of transportation mode and activity type recognition, and applies full and partial consistency principles to different trip episodes of a tour. Complementing earlier work on the evaluation of this approach at the epoch level, this paper investigates the performance of the integrated imputation at the episode level. The originally imputed data were used as reference to compare the superimposed data against validated data. Results indicate that the distribution of transportation modes and activity types are similar for these data sets. The new algorithm imputes diary data that are closer to the validated data that the results of the original algorithm.

Comparison of Raw Acceleration from the GENEA and ActiGraph™ GT3X+ Activity Monitors

Purpose: To compare raw acceleration output of the ActiGraph™ GT3X+ and GENEA activity monitors. Methods: A GT3X+ and GENEA were oscillated in an orbital shaker at frequencies ranging from 0.7 to 4.0 Hz (ten 2-min trials/frequency) on a fixed radius of 5.08 cm. Additionally, 10 participants (age = 23.8 ± 5.4 years) wore the GT3X+ and GENEA on the dominant wrist and performed treadmill walking (2.0 and 3.5 mph) and running (5.5 and 7.5 mph) and simulated free-living activities (computer work, cleaning a room, vacuuming and throwing a ball) for 2-min each. A linear mixed model was used to compare the mean triaxial vector magnitude (VM) from the GT3X+ and GENEA at each oscillation frequency. For the human testing protocol, random forest machine-learning technique was used to develop two models using frequency domain (FD) and time domain (TD) features for each monitor. We compared activity type recognition accuracy between the GT3X+ and GENEA when the prediction model was fit using one monitor and then applied to the other. Z-statistics were used to compare the proportion of accurate predictions from the GT3X+ and GENEA for each model. Results: GENEA produced significantly higher (p < 0.05, 3.5 to 6.2%) mean VM than GT3X+ at all frequencies during shaker testing. Training the model using TD input features on the GENEA and applied to GT3X+ data yielded significantly lower (p < 0.05) prediction accuracy. Prediction accuracy was not compromised when interchangeably using FD models between monitors. Conclusions: It may be inappropriate to apply a model developed on the GENEA to predict activity type using GT3X+ data when input features are TD attributes of raw acceleration.

Daily physical activity assessment with accelerometers: new insights and validation studies

The field of application of accelerometry is diverse and ever expanding. Because by definition all physical activities lead to energy expenditure, the doubly labelled water (DLW) method as gold standard to assess total energy expenditure over longer periods of time is the method of choice to validate accelerometers in their ability to assess daily physical activities. The aim of this paper was to provide a systematic overview of all recent (2007-2011) accelerometer validation studies using DLW as the reference. The PubMed Central database was searched using the following keywords: doubly or double labelled or labeled water in combination with accelerometer, accelerometry, motion sensor, or activity monitor. Limits were set to include articles from 2007 to 2011, as earlier publications were covered in a previous review. In total, 38 articles were identified, of which 25 were selected to contain sufficient new data. Eighteen different accelerometers were validated. There was a large variability in accelerometer output and their validity to assess daily physical activity. Activity type recognition has great potential to improve the assessment of physical activity-related health outcomes. So far, there is little evidence that adding other physiological measures such as heart rate significantly improves the estimation of energy expenditure.

Trip Chain's Activity Type Recognition Based on Support Vector Machine

This paper focuses on support vector machine (SVM) based trip chain's activity type recognition. First, the time-series location information of person trip is processed to obtain the trip chain elements including moving processes and activities, and the activity options are extracted from the geographic information system (GIS) around the activity sites. Second, the activity features are drawn from spatio-temporal factors of trip chain to serve as the input feature vector of classifier. A SVM based one-to-one classifier is established and the method of one-to-one classifier voting is adopted to decide the most likely activity type from the activity options. Finally, the classifiers are trained with simulation data based on the Gaussian radial basis (RBF) kernel function and the multilayer perception (MLP) kernel function respectively, and then examined by cross validation. The result shows that in the one-to-one classifying scheme, the highest and lowest right recognition rate with RBF are 99% and 62%, and the corresponding results with MLP are 97% and 54%, respectively.

Mental activity type recognition with few seconds of on-going EEG

Event Abstract Back to Event Mental activity type recognition with few seconds of on-going EEG Roman Naumov1*, Georgy Ivanitsky1 and Alexey Ivanitsky1 1 IHNA RAS, Human Higher Nervous Activity Laboratory, Russia A new method for recognition of mental task type being solved in mind is elaborated. The recognition is based on the analysis of EEG rhythms. Few seconds of EEG are enough to complete the recognition since the method utilizes an Artificial Neural Network (ANN). Method: 240 tasks of eight different types belonging to two basic thinking modes (spatial and verbal) were presented to a subject in a random order during an experiment. A total of 30 experiments with 15 healthy adult subjects were carried out. Epochs of EEG correspondent to task solving in mind were selected, inspected for artifacts and processed. Single power spectra, i.e. square modules of Fast Fourier Transform of single EEG trials, were calculated. These spectra served as an input to ANN. A Perceptron -like ANN was utilized as a learned classifier. First, the ANN was learned to recognize the task types correctly with a set of data, whose class was known. Later, the network recognized the unknown (new) data. The average duration of single EEG trials was about 10 seconds, the minimum was 2 seconds. Results:Two basic modes of thinking, i.e. verbal vs. spatial, were discriminated with confidence (average correct recognition score was about 90%). This meant that, being learned to recognize basic thinking classes with any four task types (two of them belonging to spatial thinking, and the other two – to verbal) the ANN then determines the basic class of other four task types correctly. The recognition of particular task types within one basic thinking class could also be performed, but with much less accuracy (about 66 % on average). Perfection of preprocessing and classification methods may lead to the improvement of recognition scores. The obtained results may be useful in some practical applications. Conference: Neuroinformatics 2009, Pilsen, Czechia, 6 Sep - 8 Sep, 2009. Presentation Type: Poster Presentation Topic: Electrophysiology Citation: Naumov R, Ivanitsky G and Ivanitsky A (2019). Mental activity type recognition with few seconds of on-going EEG. Front. Neuroinform. Conference Abstract: Neuroinformatics 2009. doi: 10.3389/conf.neuro.11.2009.08.062 Copyright: The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers. They are made available through the Frontiers publishing platform as a service to conference organizers and presenters. The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated. Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed. For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions. Received: 22 May 2009; Published Online: 09 May 2019. * Correspondence: Roman Naumov, IHNA RAS, Human Higher Nervous Activity Laboratory, Moscow, Russia, roman.naumov@gmail.com Login Required This action requires you to be registered with Frontiers and logged in. To register or login click here. Abstract Info Abstract The Authors in Frontiers Roman Naumov Georgy Ivanitsky Alexey Ivanitsky Google Roman Naumov Georgy Ivanitsky Alexey Ivanitsky Google Scholar Roman Naumov Georgy Ivanitsky Alexey Ivanitsky PubMed Roman Naumov Georgy Ivanitsky Alexey Ivanitsky Related Article in Frontiers Google Scholar PubMed Abstract Close Back to top Javascript is disabled. Please enable Javascript in your browser settings in order to see all the content on this page.