Human Activity Recognition Using Ambient Sensor Data

Smart support systems for the recognition of Activities of Daily Living (ADLs) can help elderly people live independently for longer improving their standard of living. Many machine learning approaches have been proposed lately for Human Activity Recognition (HAR), including elaborated networks that contain convolutional, recurrent, and attentive layers. The ubiquity of wearable devices has provided an increasing amount of time-series data that can be used for such applications in an unobtrusive manner. But there are not many studies on the performance of the attention-based Transformer model in HAR, especially not for complex activities such as ADLs. This work implements and evaluates the novel self-attention Transformer model for the classification of ADLs and compares it to the already well-established approach of recurrent Long-Short Term Memory (LSTM) networks. The proposed method is a two-level hierarchical model, in which atomic activities are initially recognized in the first step and their probability scores are extracted and utilized for the Transformer-based classification of seven more complex ADLs in the second step. The Transformer is used at the second step to classify seven ADLs. Our results show that the Transformer model reaches the same performance and even outperforms LSTM networks cleary in the subject-dependent configuration (73.36 % and 69.09 %), while relying only on attention-mechanism to depict global dependencies between input and output without the need to use any recurrence. The proposed model was tested using two different segment lengths, indicating its effectiveness in learning long-range dependencies of shorter actions in complex activities.

Accuracy of data and information on various types of land use and land cover (LULC) is crucial in watershed management planning. The development of remote sensing technology in predicting LULC has been widely used, such as the Support Vector Machine (SVM) and Random Trees (RT) methods. Both methods are part of the model of machine learning and belong to the category of supervised learning. This method can predict, study, and recognize data patterns from the results of training data. The aim of this research is the use of SVM and RT methods for the identification of LULC. Based on the SVM method analysis results, the forest area is 211.61 km2, coffee plantations are 32.92 km2, shrubs are 16.85 km2, built-up/bare land is 3.40 km2, the water body is 3.17 km2, and cloud/shadow is 6.79 km2. While the RT method obtained a forest area of 215.51 km2, coffee plantations covering an area of 29.32 km2, shrubs are covering an area of 16.32 km2, built-up/bare land surrounding an area of 7.22 km2, water body covering an area of 2.62 km2, and cloud/shadow covering an area of 3.75 km2. Overall, the percentage accuracy of the SVM method of 87%, and the RT of 86%. The success rate of modeling SVM (AUC = 0.73) and RT (AUC = 0.75) based on the results of data validation using the area under the receiver operating characteristic curve (AUC) was considered fair good.

Human activity recognition is considered a challenging task in sensor-based monitoring systems. In ambient intelligent environments, such as smart homes, collecting data from ambient sensors is useful for recognizing activities of daily living, which can then be used to provide assistance to inhabitants. Activities of daily living are composed of complex multivariable time series data that has high dimensionality, is huge in size, and is updated constantly. Thus, developing methods for analyzing time series data to extract meaningful features and specific characteristics would help solve the problem of human activity recognition. Based on the noticeable success of deep learning in the field of time series classification, we developed a model called a deep one-dimensional convolutional neural network (Deep 1d-CNN) for recognizing activities of daily living in smart homes. Our model contains several one-dimensional convolution layers coupled with max-pooling technique to learn the internal representation of time series data and automatically generate very deep features for recognizing different activity types. For the performance evaluation, we tested our deep model on the new real-life dataset, ContextAct@A4H, and the results showed that our model achieved a high F1 score (0.90). We also extended our study to show the potential energy saving in smart homes through recognizing activities of daily living. We built a recommendation system based on the activities recognized by our deep model to detect the devices that are wasting energy, and recommend the user to execute energy optimization actions. The experiment indicated that recognizing activities of daily living can result in energy savings of around 50%.

The need for a human activity recognition system arises when designing a smart home which monitors its occupants to assess their health status. In this work, a rule-based system was constructed to classify the common activities of daily living based on a hierarchical approach, using location measurements from a commercial ultrasonic sensor system. Adaptive rule application was achieved by applying contextual information from adjacent time steps using a finite-state machine. Some common static and dynamic activities of daily living were chosen as the targets for classification. The system was shown to provide comparable performance with results which have been reported for more complex alternative systems. Results reported showed a minimum classification accuracies of 87.7% for the walking activity. The deployed adaptive rule-based system provides a robust and computationally inexpensive solution for common in-situ human activity recognition purposes.

Physical activity in daily life and exercise capacity have not been assessed in patients with Crohn's disease to date. To evaluate the physical activity in daily life, exercise capacity, quality of life, and prevalence of mood disorders in patients with moderate-to-severe Crohn's disease on infliximab-induced remission and the possible associations among variables. A cross-sectional preliminary study was conducted. Twenty-six patients with Crohn's disease and 20 controls were selected. Participants underwent evaluation of physical activity in daily life (triaxial accelerometer), exercise capacity (shuttle walk test), handgrip strength, quality of life, and presence of mood disorders. The number of steps taken (7446±3081 vs 7898±2487), active time (80.6±42 vs 89.7±24.3min), shuttle walk test distance [665 (405) vs 710 (409) m] and handgrip strength [31 (15) vs 29 (20) kgf did not show any difference between the patients with Crohn's disease and the controls. The time spent lying down [95.8 (68.8) vs 60.9 (74.7) min] was greater and some domains of the quality of life were superior in the patients with Crohn's disease. No correlation was observed between the physical activity in daily life and quality of life or presence of mood disorders in patients with Crohn's disease. Patients with Crohn's disease on infliximab-induced remission, despite to more time spent lying down, they have the same level of physical activity in daily life and exercise capacity min compared with the controls.

Burnout is becoming a global pandemic jeopardizing brain health, with a huge impact on quality of life, available workforce, and the economy. Knowledge of the impact of burnout on cognition, physiology, and physical activity (PA) in daily life allows for an improved understanding of the health consequences and everyday ramifications of burnout. Twenty-eight volunteers participated in a three-day recording of daily physiology and PA, including heart rate (HR) and daily steps, with a wearable device. They filled in questionnaires screening for burnout (BBI-15), depression (BDI), and executive functions (EFs) in daily life (BRIEF-A). The subjects with burnout had more challenges in EFs, higher average HRs and lower numbers of steps in daily life than those without it. The BBI-15 scores correlated positively with the BDI scores and BRIEF-A indices and negatively with the awake HR variability (HRV) and daily steps. The metacognition index correlated negatively with the HRV. In conclusion, burnout is linked with compromised EFs along with alterations in cardiac physiology and PA in daily life. Such alterations may be easily detected with wearable devices, opening possibilities for novel biomarkers of burnout and other neuropsychiatric disorders. We suggest that physical activity and heart and brain health are intimately intertwined and that burnout interacts with each of them bidirectionally.

FLAAP: An Open Human Activity Recognition (HAR) Dataset for Learning and Finding the Associated Activity Patterns

This study aimed to investigate independent relationships of daily non-exercise life activity and leisure-time exercise volume and intensity with the prevalence of metabolic syndrome and its traits in Japanese adults. Data of 24,625 eligible subjects (12,709 men, 11,916 women) who participated in the baseline survey of the Japan Multi-Institutional Collaborative Cohort (J-MICC) Study were analyzed. Information about lifestyle characteristics was obtained from a questionnaire. Logistic regression analyses were performed to evaluate the independent associations of daily life activity as well as leisure-time exercise volume and intensity with the prevalence of metabolic syndrome and its traits by sex. Male subjects with higher daily life activity as well as with higher leisure-time exercise volume had a lower prevalence of metabolic syndrome, independently with each other. Female subjects with higher daily life activity also had a lower prevalence of metabolic syndrome. Particularly, male and female subjects with the highest daily life activity quartile showed considerably low odds ratios of 0.66 (95% CI, 0.53-0.81) and 0.64 (0.52-0.79), respectively, for low HDL-cholesterol even after the adjustment for BMI compared with the first quartile. Meanwhile, male subjects with the higher leisure-time exercise showed a quite lower prevalence of elevated triglycerides. Higher moderate-intensity exercise was more intensely associated with a lower prevalence of metabolic syndrome and some of its traits in both sexes. Our results suggest that higher daily life activity and higher moderate-intensity exercise may be independently associated with a lower risk of metabolic syndrome in Japanese adults.

The automatic identification of human activities is an important application for enabling remote home monitoring of health status in elderly and fragile people. Nowadays, commercially available wearable devices found a widespread diffusion, embedding various types of sensors that can be used for correctly recognizing activities of daily living (ADL). This work deals with the identification of ADL based on upper limb gestures from a single magnetic inertial measurement unit (MIMU) placed on the wrist. In particular, emphasis was devoted to the identification of drinking and pill intake, being direct indicators of water and medication assumption. To this aim, machine learning (ML) and deep learning (DL) models were directly compared, and single MIMU recordings were evaluated on three classification experiments. Outcomes showed that DL pipeline outperformed ML in distinguishing 14 ADL, with the best accuracy provided by the combination of gyroscope and magnetometer data (about 93%). The same configuration reached over 97% accuracy in identify drinking and pill intake among the other 12 confounding ADL. This study is a step toward the development of a robust solution for the continuous and minimally invasive monitoring of hydration behaviors and adherence to medical prescription within the home environment.Clinical relevance- This work provides evidences about the feasibility of a continuous and non-invasive monitoring of the hydration behaviors and adherence to medical prescriptions in elderly individuals.

Introduction: The impact of several dementia syndromes on activities of daily living (ADLs) has been well documented, but no study has yet investigated functional ability in posterior cortical atrophy (PCA). The primarily visual nature of deficits in this condition is likely to have a pronounced impact on ADLs. Objective: The aim of this study was to profile functional change in PCA and identify predictors of change. Method: Twenty-nine PCA patients and 25 patients with typical Alzheimer’s disease (AD) and their caregivers were included in this cross-sectional study. ADLs were assessed using the Disability Assessment for Dementia (DAD), administered to caregivers, assessing basic ADLs (e.g., eating, dressing) and instrumental ADLs (e.g., managing finances, meal preparation). The predictive utility of cognitive domains (Addenbrooke’s Cognitive Examination), behavioural impairment (Cambridge Behavioural Inventory-Revised) and demographic variables on ADL ability was also examined. Results: PCA patients showed significantly reduced total ADL scores compared to AD patients (medium effect size, d = –0.7; p < 0.05), with significantly more impairment on basic ADLs (large effect size, d = –0.8; p < 0.05) but similar impairment on instrumental ADLs (medium effect size, d = –0.5; p > 0.05). A model combining patient mood, disinhibition, apathy, symptom duration, and memory and attention/orientation scores explained the variance of scores in functional decline (61.2%), but the key factor predicting ADL scores was attention/orientation (p = 0.048). Conclusion: This study shows the profound impact of PCA on ADLs and factors underpinning patients’ disability. Attention/orientation deficits were found to correlate and contribute to variance in ADL scores. Future work to develop tailored interventions to manage ADL impairment in PCA should take these findings into account.

Human Activity Recognition (HAR) has become one of the most important research fields to achieve real-time monitoring of human activities for timely decision making in various applications like fall detection, elderly care etc. Now-a-days, most people use smartphones which come with various embedded inertial sensors like accelerometer and gyroscope to monitor acceleration and angular velocity. These smartphone-based sensors have proven to be cost-effective solution in identification of activities belonging to ADL (Activities of Daily Living). Various Machine Learning, Deep learning and hybrid models have been proposed and implemented for HAR. This paper also proposes various hybrid deep learning approaches which combine Deep Neural Networks with other models like LSTM (Long Short Term Memory) Model and GRU (Gated Recurrent Unit) for effective classification of engineered features from CNN (Convolutional Neural Network) Model. A novel architecture that integrates CNN with Random Forest Classifier (DeepCNN-RF) is proposed to add randomness to the model. The proposed models have been tested on publicly available HAR Datasets like UCI HAR and WISDM Activity Recognition Datasets. Experimental results show that the hybrid models outperform the state-of-the-art data mining, machine learning techniques in UCI HAR and WISDM with an overall maximum accuracy of 97.77% and 98.2% respectively.

Existing machine learning algorithms for human information inference are typically data-driven models trained on carefully labeled datasets. Given the significant labeling effort, traditional pure data-driven approaches are challenging to implement for emerging smart applications requiring long-term finer-grained information. Taking activities of daily life (ADL) tracking for elders as an example, prior work mostly focused on context-level information learning such as cooking and cleaning. [8]. However, new applications such as evaluating elders' cognitive impairments progress by tracking their ADL engagement requires finer-grained, i.e., action-level information [7]. In practice, labeling the day-length data at such granularity can be very expensive and requires a lot of human efforts [9]. My research focuses on the inference problems in the scope of human physical condition monitoring and activity recognition with limited labeled data. To alleviate the effort of labeling large amounts of data, prior works on semi-supervised learning combine a small amount of labeled data with a large amount of unlabeled data to train the model. However, as the label granularity (number of classes) increasing, the difficulty to distinguish nuance distinctions between finer-grained classes escalates as well. This makes training a robust semi-supervised model for finer-grained classification with less labels difficult if not impossible. Fortunately, coarse-grained (context-level) labels is usually available or cheaper to obtain in practice. In this case, the multi-granularity hierarchy between finer and coarse labels follows the aggregation relation defined in [5]. This hierarchical relation can be leveraged in the tasks of inferring finer-grained information. In addition, it is illustrated by the previous study that co-located multi modality sensing systems capture complementary aspects of the same event [6]. The research question I focus on is how to infer finer-grained human information with coarse-grained labeled data leveraging complementary multi-modal sensing? I target three directions: 1) a cross-granularity semi-supervised setting: how to utilize coarse-grained labeled data with a small amount of finer-grained labeled data to infer finer-grained human information, 2) cross-granularity relationship learning: how to learn the multi-granularity class hierarchy from data and further help the finer-grained human information acquisition, 3) enhancing inference granularity by leveraging multi-model sensing: how to leverage the complimentary co-located multiple sensing modalities to accurately infer finer-grained human information?

To compare a weight training alone treatment (WT) to an innovative WT plus education treatment (WT + ED) about the use of strength-training gains when performing activities of daily living (ADL) with respect to their effects on ADL self-efficacy and performance. Twenty-three men and 41 women (mean age = 74.4 +/- 3.7 yr) were randomly assigned to WT or WT + ED. Both groups performed 12 wk (two sessions per week) of WT targeting eight major muscle groups. WT + ED received behavioral training and associated written materials emphasizing the link between WT and ADL. WT received a placebo educational intervention. Baseline and posttest measures were collected for self-efficacy for performing eight lab-based ADL tasks and performance of the eight ADL tasks. A manipulation check compared participants' knowledge of ADL that might be improved through WT. The WT + ED treatment listed more ADL that could be improved with WT and had greater posttest self-efficacy for performing the ADL lab tasks than the WT treatment. Greater ADL self-efficacy did not translate into better ADL performance. A targeted educational intervention can help older adults generalize the benefits and confidence obtained through WT to their performance of ADL. Further research is needed to determine the behavioral and psychosocial impact of enhanced ADL self-efficacy on older adults.

Subjective impairments in COPD are associated with dynamic hyperinflation during daily life activities

Recently, a significant amount of literature concerning machine learning techniques has focused on automatic recognition of activities performed by people. The main reason for this considerable interest is the increasing availability of devices able to acquire signals which, if properly processed, can provide information about human activities of daily living (ADL). The recognition of human activities is generally performed by machine learning techniques that process signals from wearable sensors and/or cameras appropriately arranged in the environment. Whatever the type of sensor, activities performed by human beings have a strong subjective characteristic that is related to different factors, such as age, gender, weight, height, physical abilities, and lifestyle. Personalization models have been studied to take into account these subjective factors and it has been demonstrated that using these models, the accuracy of machine learning algorithms can be improved. In this work we focus on the recognition of human activities using signals acquired by the accelerometer embedded in a smartphone. The contributions of this research are mainly three. A first contribution is the definition of a clear validation model that takes into account the problem of personalization and which thus makes it possible to objectively evaluate the performances of machine learning algorithms. A second contribution is the evaluation, on three different public datasets, of a personalization model which considers two aspects: the similarity between people related to physical aspects (age, weight, and height) and similarity related to intrinsic characteristics of the signals produced by these people when performing activities. A third and last contribution is the development of a personalization model that considers both the physical and signal similarities. The experiments show that the employment of personalization models improves, on average, the accuracy, thus confirming the soundness of the approach and paving the way for future investigations on this topic.