No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Intelligent Behavioral Health Analysis in Youth Using Real-Time Smartphone and Wearable Data
Abstract
Wearable devices have sensors that can be used for collecting motion data and provide objective measures of sleep and physical activity. Smartphones are being used increasingly contributing to users’ screen viewing time and hence be capable of providing accurate objective measures for the same. Disruption in the key lifestyle patterns like sleep, physical activity, and screen viewing may result in behavioral health impairments as these correspond to the utmost engaged times during the day. This work uses real-time features of physical activity, sleeping habits, and screen viewing to assess the behavioral health of 24 youths. Data is acquired from participants’ smartphones and smartwatches and behavioral health scores are evaluated based on the extracted features. Three statistical feature filtering methods have been used for feature selection, viz. correlation attribute evaluator, information gain, and gain ratio. The features selected by these attribute evaluators have been used to train supervised learning algorithms including four baseline machine learning algorithms and two deep learning algorithms. The best performance is reported by random forest and convolutional neural network with the performance accuracy of 93.06% and 96.77%, respectively, for the prediction of behavioral health.
ResearchGate has not been able to resolve any citations for this publication.
With the progressive increase of stress, anxiety and depression in working and living environment, mental health assessment becomes an important social interaction research topic. Generally, clinicians evaluate the psychology of participants through an effective psychological evaluation and questionnaires. However, these methods suffer from subjectivity and memory effects. In this paper, a new multi- sensing wearable device has been developed and applied in self-designed psychological tests. Speech under different emotions as well as behavior signals are captured and analyzed. The mental state of the participants is objectively assessed through a group of psychological questionnaires. In particular, we propose an attention-based block deep learning architecture within the device for multi-feature classification and fusion analysis. This enables the deep learning architecture to autonomously train to obtain the optimum fusion weights of different domain features. The proposed attention-based architecture has led to improving performance compared with direct connecting fusion method. Experimental studies have been carried out in order to verify the effectiveness and robustness of the proposed architecture. The obtained results have shown that the wearable multi-sensing devices equipped with the attention-based block deep learning architecture can effectively classify mental state with better performance.
Objective:
We aimed to develop a machine learning algorithm to screen for depression and assess severity based on data from wearable devices.
Methods:
We used a wearable device that calculates steps, energy expenditure, body movement, sleep time, heart rate, skin temperature, and ultraviolet light exposure. Depressed patients and healthy volunteers wore the device continuously for the study period. The modalities were compared hourly between patients and healthy volunteers. XGBoost was used to build machine learning models and 10-fold cross-validation was applied for the validation.
Results:
Forty-five depressed patients and 41 healthy controls participated, creating a combined 5,250 days' worth of data. Heart rate, steps, and sleep were significantly different between patients and healthy volunteers in some comparisons. Similar differences were also observed longitudinally when patients' symptoms improved. Based on seven days' data, the model identified symptomatic patients with 0.76 accuracy and predicted Hamilton Depression Rating Scale-17 scores with a 0.61 correlation coefficient. Skin temperature, sleep time-related features, and the correlation of those modalities were the most significant features in machine learning.
Limitations:
The small number of subjects who participated in this study may have weakened the statistical significance of the study. There are differences in the demographic data among groups although we performed a correction for multiple comparisons. Validation in independent datasets was not performed, although 10-fold cross validation with the internal data was conducted.
Conclusion:
The results indicated that utilizing wearable devices and machine learning may be useful in identifying depression as well as assessing severity.
Understanding how people use technology remains important, particularly when measuring the impact this might have on individuals and society. However, despite a growing body of resources that can quantify smartphone use, research within psychology and social science overwhelmingly relies on self-reported assessments. These have yet to convincingly demonstrate an ability to predict objective behavior. Here, and for the first time, we compare a variety of smartphone use and ‘addiction’ scales with objective behaviors derived from Apple's Screen Time application. While correlations between psychometric scales and objective behavior are generally poor, single estimates and measures that attempt to frame technology use as habitual rather than ‘addictive’ correlate more favorably with subsequent behavior. We conclude that existing self-report instruments are unlikely to be sensitive enough to accurately predict basic technology use related behaviors. As a result, conclusions regarding the psychological impact of technology are unreliable when relying solely on these measures to quantify typical usage.
In this paper, we propose a mobile-based context-aware acute stress prediction system (CASP) that predicts a user’s stress status based on their current contextual data. The system consists of a context-aware stress prediction algorithm, and an early stage stress intervention method. In the learning phase, the context-aware stress detection algorithm uses ECG signals to identify the user’s stress status. With the aid of machine learning algorithms and cloud computing services, the stress prediction algorithm produces adaptive and personalized prediction models based on the user’s context gathered from their smartphone. The prediction models are able to adapt the changing nature of both the user’s stress status and the surrounding environment. Our evaluation results show that the CASP system is able to predict the stress status of a user using the current contextual data with an average accuracy of 78.3% as measured from ground truth data collected using biofeedback sensors.
The increasing use of electronic forms of communication presents new opportunities in the study of mental health, including the ability to investigate the manifestations of psychiatric diseases unobtrusively and in the setting of patients' daily lives. A pilot study to explore the possible connections between bipolar affective disorder and mobile phone usage was conducted. In this study, participants were provided a mobile phone to use as their primary phone. This phone was loaded with a custom keyboard that collected metadata consisting of keypress entry time and accelerometer movement. Individual character data with the exceptions of the backspace key and space bar were not collected due to privacy concerns. We propose an end-to-end deep architecture based on late fusion, named DeepMood, to model the multi-view metadata for the prediction of mood scores. Experimental results show that 90.31% prediction accuracy on the depression score can be achieved based on session-level mobile phone typing dynamics which is typically less than one minute. It demonstrates the feasibility of using mobile phone metadata to infer mood disturbance and severity.
The rapid growth of smartphone in recent years has resulted in many syndromes. Most of these syndromes are caused by excessive use of smartphone. In addition, people who tends to use smartphone excessively are also likely to have smartphone addiction. In this paper, we presented the system architecture for e-Health system. Not only we used the architecture for our smartphone addiction recognition system, but we also pointed out important benefits of the system architecture, which also can be adopted by other system. Later on, we presented a development of the classification model for recognizing likelihood of having smartphone addiction. We trained the classification model based on data retrieved from subjects’ smartphone. The result showed that the best model can correctly classify the instance up to 78%.
This piece of research is situated in the domain of multi-modal analytics. New commercial off the shelf wearables, such as smartwatches or wristbands, are becoming popular and increasingly used for fitness and wellness in a new trend known as the quantified-self movement. The sensors included in these devices (e.g. accelerometer, heart rate) in conjunction with data analytics algorithms are used to provide information such as steps walked, calories consumed, etc. The main goal of this piece of research is to check if new wearable technologies could be used to estimate sleep indicators in an automatic way. The available medical literature proposes several sleep-related features and methods to calculate them involving direct user observation, interviews or specific medical instrumentation. Off the shelf wearable vendors also provide some sleep indicators, such as the sleep duration, the number of awakes or the time to fall asleep. Taking as a reference the results and methods described in the medical literature and the data available in commercial off the shelf wearables, we propose new sleep indicators offering a greater interpretative value: sleep quality, sleepiness level, chronotype. The results obtained after initial experiments demonstrate the feasibility of this approach to be applied in real contexts. Eventually, we plan to apply these solutions to support educational scenarios related to self-regulated learning and teaching support.
Background:
The importance of sleep is paramount to health. Insufficient sleep can reduce physical, emotional, and mental well-being and can lead to a multitude of health complications among people with chronic conditions. Physical activity and sleep are highly interrelated health behaviors. Our physical activity during the day (ie, awake time) influences our quality of sleep, and vice versa. The current popularity of wearables for tracking physical activity and sleep, including actigraphy devices, can foster the development of new advanced data analytics. This can help to develop new electronic health (eHealth) applications and provide more insights into sleep science.
Objective:
The objective of this study was to evaluate the feasibility of predicting sleep quality (ie, poor or adequate sleep efficiency) given the physical activity wearable data during awake time. In this study, we focused on predicting good or poor sleep efficiency as an indicator of sleep quality.
Methods:
Actigraphy sensors are wearable medical devices used to study sleep and physical activity patterns. The dataset used in our experiments contained the complete actigraphy data from a subset of 92 adolescents over 1 full week. Physical activity data during awake time was used to create predictive models for sleep quality, in particular, poor or good sleep efficiency. The physical activity data from sleep time was used for the evaluation. We compared the predictive performance of traditional logistic regression with more advanced deep learning methods: multilayer perceptron (MLP), convolutional neural network (CNN), simple Elman-type recurrent neural network (RNN), long short-term memory (LSTM-RNN), and a time-batched version of LSTM-RNN (TB-LSTM).
Results:
Deep learning models were able to predict the quality of sleep (ie, poor or good sleep efficiency) based on wearable data from awake periods. More specifically, the deep learning methods performed better than traditional logistic regression. “CNN had the highest specificity and sensitivity, and an overall area under the receiver operating characteristic (ROC) curve (AUC) of 0.9449, which was 46% better as compared with traditional logistic regression (0.6463).
Conclusions:
Deep learning methods can predict the quality of sleep based on actigraphy data from awake periods. These predictive models can be an important tool for sleep research and to improve eHealth solutions for sleep.
Importance:
Sleep is vital to children's biopsychosocial development. Inadequate sleep quantity and quality is a public health concern with an array of detrimental health outcomes. Portable mobile and media devices have become a ubiquitous part of children's lives and may affect their sleep duration and quality.
Objective:
To conduct a systematic review and meta-analysis to examine whether there is an association between portable screen-based media device (eg, cell phones and tablet devices) access or use in the sleep environment and sleep outcomes.
Data sources:
A search strategy consisting of gray literature and 24 Medical Subject Headings was developed in Ovid MEDLINE and adapted for other databases between January 1, 2011, and June 15, 2015. Searches of the published literature were conducted across 12 databases. No language restriction was applied.
Study selection:
The analysis included randomized clinical trials, cohort studies, and cross-sectional study designs. Inclusion criteria were studies of school-age children between 6 and 19 years. Exclusion criteria were studies of stationary exposures, such as televisions or desktop or personal computers, or studies investigating electromagnetic radiation.
Data extraction and synthesis:
Of 467 studies identified, 20 cross-sectional studies were assessed for methodological quality. Two reviewers independently extracted data.
Main outcomes and measures:
The primary outcomes were inadequate sleep quantity, poor sleep quality, and excessive daytime sleepiness, studied according to an a priori protocol.
Results:
Twenty studies were included, and their quality was assessed. The studies involved 125 198 children (mean [SD] age, 14.5 [2.2] years; 50.1% male). There was a strong and consistent association between bedtime media device use and inadequate sleep quantity (odds ratio [OR], 2.17; 95% CI, 1.42-3.32) (P < .001, I2 = 90%), poor sleep quality (OR, 1.46; 95% CI, 1.14-1.88) (P = .003, I2 = 76%), and excessive daytime sleepiness (OR, 2.72; 95% CI, 1.32-5.61) (P = .007, I2 = 50%). In addition, children who had access to (but did not use) media devices at night were more likely to have inadequate sleep quantity (OR, 1.79; 95% CI, 1.39-2.31) (P < .001, I2 = 64%), poor sleep quality (OR, 1.53; 95% CI, 1.11-2.10) (P = .009, I2 = 74%), and excessive daytime sleepiness (OR, 2.27; 95% CI, 1.54-3.35) (P < .001, I2 = 24%).
Conclusions and relevance:
To date, this study is the first systematic review and meta-analysis of the association of access to and the use of media devices with sleep outcomes. Bedtime access to and use of a media device were significantly associated with the following: inadequate sleep quantity, poor sleep quality, and excessive daytime sleepiness. An integrated approach among teachers, health care professionals, and parents is required to minimize device access at bedtime, and future research is needed to evaluate the influence of the devices on sleep hygiene and outcomes.
Continuous exposure to stress is harmful for mental and physical health, but to combat stress, one should first detect it. In this paper we propose a method for continuous detection of stressful events using data provided from a commercial wrist device. The method consists of three machine-learning components: a laboratory stress detector that detects short-term stress every 2 minutes; an activity recognizer that continuously recognizes user's activity and thus provides context information; and a context-based stress detector that exploits the output of the laboratory stress detector and the user's context in order to provide the final decision on 20 minutes interval. The method was evaluated in a laboratory and a real-life setting. The accuracy on 55 days of real-life data, for a 2-class problem, was 92%. The method is currently being integrated in a smartphone application for managing mental health and well-being.
As the population increases in the world, the ratio of health carers is rapidly decreasing. Therefore, there is an urgent need to create new technologies to monitor the physical and mental health of people during their daily life. In particular, negative mental states like depression and anxiety are big problems in modern societies, usually due to stressful situations during everyday activities including work. This paper presents a machine learning approach for stress detection on people using wearable physiological sensors with the final aim of improving their quality of life. The presented technique can monitor the state of the subject continuously and classify it into ”stressful” or ”non-stressful” situations. Our classification results show that this method is a good starting point towards real-time stress detection.
Background:
The benefits of physical activity are well documented, but scalable programs to promote activity are needed. Interventions that assign tailored and dynamically adjusting goals could effect significant increases in physical activity but have not yet been implemented at scale.
Objective:
Our aim was to examine the effectiveness of an open access, Internet-based walking program that assigns daily step goals tailored to each participant.
Methods:
A two-arm, pragmatic randomized controlled trial compared the intervention to no treatment. Participants were recruited from a workplace setting and randomized to a no-treatment control (n=133) or to treatment (n=132). Treatment participants received a free wireless activity tracker and enrolled in the walking program, Walkadoo. Assessments were fully automated: activity tracker recorded primary outcomes (steps) without intervention by the participant or investigators. The two arms were compared on change in steps per day from baseline to follow-up (after 6 weeks of treatment) using a two-tailed independent samples t test.
Results:
Participants (N=265) were 66.0% (175/265) female with an average age of 39.9 years. Over half of the participants (142/265, 53.6%) were sedentary (<5000 steps/day) and 44.9% (119/265) were low to somewhat active (5000-9999 steps/day). The intervention group significantly increased their steps by 970 steps/day over control (P<.001), with treatment effects observed in sedentary (P=.04) and low-to-somewhat active (P=.004) participants alike.
Conclusions:
The program is effective in increasing daily steps. Participants benefited from the program regardless of their initial activity level. A tailored, adaptive approach using wireless activity trackers is realistically implementable and scalable.
Trial registration:
Clinicaltrials.gov NCT02229409, https://clinicaltrials.gov/ct2/show/NCT02229409 (Archived by WebCite at http://www.webcitation.org/6eiWCvBYe).
The random forest algorithm, proposed by L. Breiman in 2001, has been
extremely successful as a general-purpose classification and regression method.
The approach, which combines several randomized decision trees and aggregates
their predictions by averaging, has shown excellent performance in settings
where the number of variables is much larger than the number of observations.
Moreover, it is versatile enough to be applied to large-scale problems, is
easily adapted to various ad-hoc learning tasks, and returns measures of
variable importance. The present article reviews the most recent theoretical
and methodological developments for random forests. Emphasis is placed on the
mathematical forces driving the algorithm, with special attention given to the
selection of parameters, the resampling mechanism, and variable importance
measures. This review is intended to provide non-experts easy access to the
main ideas.
Neuropsychiatric disorders are the leading cause of disability worldwide and there is no gold standard currently available for the measurement of mental health. This issue is exacerbated by the fact that the information physicians use to diagnose these disorders is episodic and often subjective. Current methods to monitor mental health involve the use of subjective DSM-5 guidelines, and advances in EEG and video monitoring technologies have not been widely adopted due to invasiveness and inconvenience. Wearable technologies have surfaced as a ubiquitous and unobtrusive method for providing continuous, quantitative data about a patient. Here, we introduce PRISM — Passive, Real-time Information for Sensing Mental Health. This platform integrates motion, light and heart rate data from a smart watch application with user interactions and text entries from a web application. We have demonstrated a proof of concept by collecting preliminary data through a pilot study of 13 subjects. We have engineered appropriate features and applied both unsupervised and supervised learning to develop models that are predictive of user-reported ratings of their emotional state, demonstrating that the data has the potential to be useful for evaluating mental health. This platform could allow patients and clinicians to leverage continuous streams of passive data for early and accurate diagnosis as well as constant monitoring of patients suffering from mental disorders.
Smartphone usage has tremendously increased and most users keep their smartphones close throughout the day. Smartphones have a broad variety of sensors, that could automatically map and track the user's life and behaviour. In this work we investigate whether automatically collected smartphone usage and sensor data can be employed to predict the experienced stress levels of a user using a customized brief version of the Perceived Stress Scale (PSS). To that end we have conducted a user study in which smartphone data and stress (as measured by the PSS seven times a day) were recorded for two weeks. We found significant correlations between stress scores and smartphone usage as well as sensor data, pointing to innovative ways for automatic stress measurements via smartphone technology. Stress is a prevalent risk factor for multiple diseases. Thus accurate and efficient prediction of stress levels could provide means for targeted prevention and intervention.
Decision tree methodology is a commonly used data mining method for establishing classification systems based on multiple covariates or for developing prediction algorithms for a target variable. This method classifies a population into branch-like segments that construct an inverted tree with a root node, internal nodes, and leaf nodes. The algorithm is non-parametric and can efficiently deal with large, complicated datasets without imposing a complicated parametric structure. When the sample size is large enough, study data can be divided into training and validation datasets. Using the training dataset to build a decision tree model and a validation dataset to decide on the appropriate tree size needed to achieve the optimal final model. This paper introduces frequently used algorithms used to develop decision trees (including CART, C4.5, CHAID, and QUEST) and describes the SPSS and SAS programs that can be used to visualize tree structure.
Smartphone adoption has increased significantly and, with the increase in smartphone capabilities, this means that users can access the Internet, communicate, and entertain themselves anywhere and anytime. However, there is growing evidence of problematic use of smartphones that impacts both social and heath aspects of users' lives. Currently, assessment of overuse or problematic use depends on one-time, self-reported behavioral information about phone use. Due to the known issues with self-reports in such types of assessments, we explore an automated, objective and repeatable approach for assessing problematic usage. We collect a wide range of phone usage data from smartphones, identify a number of usage features that are relevant to this assessment, and build detection models based on Adaboost with machine learning algorithms automatically detecting problematic use. We found that the number of apps used per day, the ratio of SMSs to calls, the number of event-initiated sessions, the number of apps used per event initiated session, and the length of non-event-initiated sessions are useful for detecting problematic usage. With these, a detection model can identify users with problematic usage with 89.6% accuracy (F-score of .707).
How we feel is greatly influenced by how well we sleep. Emerging quantified-self apps and wearable devices allow people to measure and keep track of sleep duration, patterns and quality. However, these approaches are intrusive, placing a burden on the users to modify their daily sleep related habits in order to gain sleep data; for example, users have to wear cumbersome devices (e.g., a headband) or inform the app when they go to sleep and wake up. In this paper, we present a radically different approach for measuring sleep duration based on a novel best effort sleep (BES) model. BES infers sleep using smartphones in a completely unobtrusive way - that is, the user is completely removed from the monitoring process and does not interact with the phone beyond normal user behavior. A sensor-based inference algorithm predicts sleep duration by exploiting a collection of soft hints that tie sleep duration to various smartphone usage patterns (e.g., the time and length of smartphone usage or recharge events) and environmental observations (e.g., prolonged silence and darkness). We perform quantitative and qualitative comparisons between two smartphone only approaches that we developed (i.e., BES model and a sleep-with-the-phone approach) and two popular commercial wearable systems (Le., the Zeo headband and Jawbone wristband). Results from our one-week 8-person study look very promising and show that the BES model can accurately infer sleep duration (± 42 minutes) using a completely "hands off" approach that can cope with the natural variation in users' sleep routines and environments.
The naive Bayes classifier greatly simplify learn-ing by assuming that features are independent given class. Although independence is generally a poor assumption, in practice naive Bayes often competes well with more sophisticated classifiers. Our broad goal is to understand the data character-istics which affect the performance of naive Bayes. Our approach uses Monte Carlo simulations that al-low a systematic study of classification accuracy for several classes of randomly generated prob-lems. We analyze the impact of the distribution entropy on the classification error, showing that low-entropy feature distributions yield good per-formance of naive Bayes. We also demonstrate that naive Bayes works well for certain nearly-functional feature dependencies, thus reaching its best performance in two opposite cases: completely independent features (as expected) and function-ally dependent features (which is surprising). An-other surprising result is that the accuracy of naive Bayes is not directly correlated with the degree of feature dependencies measured as the class-conditional mutual information between the fea-tures. Instead, a better predictor of naive Bayes ac-curacy is the amount of information about the class that is lost because of the independence assump-tion.
An approach to the construction of classifiers from imbalanced datasets is
described. A dataset is imbalanced if the classification categories are not
approximately equally represented. Often real-world data sets are predominately
composed of "normal" examples with only a small percentage of "abnormal" or
"interesting" examples. It is also the case that the cost of misclassifying an
abnormal (interesting) example as a normal example is often much higher than
the cost of the reverse error. Under-sampling of the majority (normal) class
has been proposed as a good means of increasing the sensitivity of a classifier
to the minority class. This paper shows that a combination of our method of
over-sampling the minority (abnormal) class and under-sampling the majority
(normal) class can achieve better classifier performance (in ROC space) than
only under-sampling the majority class. This paper also shows that a
combination of our method of over-sampling the minority class and
under-sampling the majority class can achieve better classifier performance (in
ROC space) than varying the loss ratios in Ripper or class priors in Naive
Bayes. Our method of over-sampling the minority class involves creating
synthetic minority class examples. Experiments are performed using C4.5, Ripper
and a Naive Bayes classifier. The method is evaluated using the area under the
Receiver Operating Characteristic curve (AUC) and the ROC convex hull strategy.
Because of the quick development and widespread use of mobile phones, and their vast effect on communication and interactions, it is important to study possible negative health effects of mobile phone exposure. The overall aim of this study was to investigate whether there are associations between psychosocial aspects of mobile phone use and mental health symptoms in a prospective cohort of young adults.
The study group consisted of young adults 20-24 years old (n = 4156), who responded to a questionnaire at baseline and 1-year follow-up. Mobile phone exposure variables included frequency of use, but also more qualitative variables: demands on availability, perceived stressfulness of accessibility, being awakened at night by the mobile phone, and personal overuse of the mobile phone. Mental health outcomes included current stress, sleep disorders, and symptoms of depression. Prevalence ratios (PRs) were calculated for cross-sectional and prospective associations between exposure variables and mental health outcomes for men and women separately.
There were cross-sectional associations between high compared to low mobile phone use and stress, sleep disturbances, and symptoms of depression for the men and women. When excluding respondents reporting mental health symptoms at baseline, high mobile phone use was associated with sleep disturbances and symptoms of depression for the men and symptoms of depression for the women at 1-year follow-up. All qualitative variables had cross-sectional associations with mental health outcomes. In prospective analysis, overuse was associated with stress and sleep disturbances for women, and high accessibility stress was associated with stress, sleep disturbances, and symptoms of depression for both men and women.
High frequency of mobile phone use at baseline was a risk factor for mental health outcomes at 1-year follow-up among the young adults. The risk for reporting mental health symptoms at follow-up was greatest among those who had perceived accessibility via mobile phones to be stressful. Public health prevention strategies focusing on attitudes could include information and advice, helping young adults to set limits for their own and others' accessibility.
Wearable devices equipped with sensors popularly used for health monitoring are capable of accumulating motion data providing objective measures of various physical activity and sleep attributes. Also, smartphone usage has grown to an extent where phones have become an integral part of lifestyle contributing to users’ screen viewing time. Behavior and behavioral attributes of individuals’ personal characteristics are significant components of lifestyle of which sleep, physical activity and screen viewing correspond to the most occupied time throughout the day and among the major lifestyle patterns affecting overall health. This study aims to assess sleep quality and behavioral health from wearables and smartphones using an Adaptive Neuro-Fuzzy Inference System (ANFIS). Real time physical activity and sleep data have been collected from users’ smartwatches. A smartphone application is built to collect real time smartphone usage data. Sleep quality indicator (SleepQual) for assessment of daily sleep quality is calculated using sleep attributes collected from smartwatches. Correlation of SleepQual is evaluated with physical activity attributes and smartphone usage attributes using Pearson’s correlation. Highly correlated attributes along with sleep attributes are used to train the ANFIS model for sleep quality assessment. A novel behavioral health indicator (B. Health) is proposed which is evaluated using real time physical activity, sleep and screen time data. Attributes are ranked on the basis of Pearson’s correlation with B. Health to identify the most important contributors to behavioral health. Top ranked features are selected to train the ANFIS model for behavioral health assessment. Systematic Minority Oversampling Technique has been used for data augmentation. The ANFIS model achieves an accuracy of 91.69% for sleep quality assessment and 85.79% for behavioral health assessment.
Many nations have imposed lockdowns due to the COVID-19 pandemic as a measure to prevent the spread of disease among its population. These lockdowns have confined people at their homes which is leading them to use digital technologies such as Internet, social media, smartphones, more than ever before. The problematic use of these digital technologies may impact their mental and emotional health. This chapter discusses the role of machine learning to assess addiction to various digital technologies and its impact on mental and emotion health and on sleep quality during the COVID-19 pandemic. Three case studies are provided to demonstrate how machine learning can be used to assess these addictions and related disorders during the pandemic. Gaussian mixture clustering is implemented to group people with similar Twitter usage patterns to identify addictive Twitter usage during the pandemic. The results convey that 11.71% of users show addictive Twitter usage patterns and 4.05% of users show highly addictive Twitter usage patterns while 2.70% of users show dangerously addictive usage patterns. “Sadness” and “anger” are the dominating emotions among these users in contrast to “happiness” which is the dominating emotion among non-addictive users. A similar approach is used to cluster students with similar smartphone usage patterns and nomophobia scores to identify nomophobic behavior during the pandemic. The results show that 4.5% of students are at extremely high risk whereas 73% of students are at high risk. A review of studies identifies the emergence of machine learning for assessment of mental and emotional health during the COVID-19 pandemic. A case study on sleep quality assessment using data from wearable sensors convey that sleep quality of students has been reduced significantly during the pandemic with a maximum decrease of 90.90%.
Wearable devices such as smartwatches, wristbands, GPS shoes are increasingly used for fitness and wellness as they allow users to monitor their daily health. These devices have sensors for accumulating user activity data. Clinical actigraph devices fall in the category of wearable devices worn on the wrist determined to estimate sleep parameters by recording movements during sleep. This study aims to predict sleep quality from wearable sensors using deep learning techniques. Three sleep indicators are proposed which are calculated using the data collected automatically from wearable devices. These sleep indicators are Daily Sleep Quality, Weekly Sleep Quality, and Sleep Consistency. Two deep learning models namely Convolution Neural Network (CNN) and Multilayer Perceptron (MLP) have been implemented to predict sleep quality on the basis of the proposed indicators. Two datasets have been used to validate the work proposed in this study which include a dataset comprising sleep parameters using commercial wearable devices and another dataset consisting of sleep data using clinical actigraph device. Systematic Minority Oversampling Technique has been applied for data augmentation with the intent to increase data instances and to alleviate class imbalance. CNN is observed to outperform MLP in predicting sleep quality with the highest accuracy of 97.30%. This study also evaluates the worth of each sleep attribute using Information Gain algorithm in order to identify the most important attributes which contribute to the sleep quality. It has been concluded that in bed awake percentage contributes maximum to the Daily Sleep Quality, average sleep efficiency contributes maximum to the Weekly Sleep Quality and standard deviation of midpoint of in bed and out of bed times contributes maximum to the Sleep Consistency.
Background
This paper aims to synthesise the literature on machine learning (ML) and big data applications for mental health, highlighting current research and applications in practice.
Methods
We employed a scoping review methodology to rapidly map the field of ML in mental health. Eight health and information technology research databases were searched for papers covering this domain. Articles were assessed by two reviewers, and data were extracted on the article's mental health application, ML technique, data type, and study results. Articles were then synthesised via narrative review.
Results
Three hundred papers focusing on the application of ML to mental health were identified. Four main application domains emerged in the literature, including: (i) detection and diagnosis; (ii) prognosis, treatment and support; (iii) public health, and; (iv) research and clinical administration. The most common mental health conditions addressed included depression, schizophrenia, and Alzheimer's disease. ML techniques used included support vector machines, decision trees, neural networks, latent Dirichlet allocation, and clustering.
Conclusions
Overall, the application of ML to mental health has demonstrated a range of benefits across the areas of diagnosis, treatment and support, research, and clinical administration. With the majority of studies identified focusing on the detection and diagnosis of mental health conditions, it is evident that there is significant room for the application of ML to other areas of psychology and mental health. The challenges of using ML techniques are discussed, as well as opportunities to improve and advance the field.
Sleep is the most important aspect of healthy and active living. The right amount of sleep at the right time helps an individual to protect his or her physical, mental, and cognitive health and maintain his or her quality of life. The most durative of the Activities of Daily Living (ADL), sleep has a major synergic influence on a person’s fuctional, behavioral, and cognitive health. A deep understanding of sleep behavior and its relationship with its physiological signals, and contexts (such as eye or body movements), is necessary to design and develop a robust intelligent sleep monitoring system. In this article, we propose an intelligent algorithm to detect the microscopic states of sleep that fundamentally constitute the components of good and bad sleeping behaviors and thus help shape the formative assessment of sleep quality. Our initial analysis includes the investigation of several classification techniques to identify and correlate the relationship of microscopic sleep states with overall sleep behavior. Subsequently, we also propose an online algorithm based on change point detection to process and classify the microscopic sleep states. We also develop a lightweight version of the proposed algorithm for real-time sleep monitoring, recognition, and assessment at scale. For a larger deployment of our proposed model across a community of individuals, we propose an active-learning-based methodology to reduce the effort of ground-truth data collection and labeling. Finally, we evaluate the performance of our proposed algorithms on real data traces and demonstrate the efficacy of our models for detecting and assessing the fine-grained sleep states beyond an individual.
It has been reported that depression can be detected by electrophysiological signals. However, few studies investigate how to daily monitor patient’s electrophysiological signals through a more convenient way for a doctor, especially on the monitoring of electroencephalogram (EEG) signals for depression diagnosis. Since a person’s mental state and physiological state are changing over time, the most insured diagnosis of depression requires doctors to collect and analyze subject’s EEG signals every day until two weeks for the clinical practice. In this work, we designed a real-time depression monitoring system to capture the user’s EEG data by a wearable device and to perform real-time signal filtering, artifacts removal and power spectrum visualization, which could be combined with psychological test scales as an auxiliary diagnosis. In addition to collecting the resting EEG signals for real-time analysis or diagnosis of depression, we also introduced an external audio stimulus paradigm to further make a detection of depression. Through the machine learning method, system can give a credible probability of depression under each stimulus as a user’s self-rating score from continuous EEG data. EEG signals collected from 81 early-onset patients and 89 normal controls are used to build the final classification model and to verify the practical performance.
The overuse of smartphones is increasingly becoming a social problem. In this paper, we analyze smartphone overuse levels, according to emotion, by examining brainwaves and deep learning. We assessed the asymmetry power with respect to theta, alpha, beta, gamma, and total brainwave activity in 11 lobes. The Deep Belief Network (DBN) was used as the deep learning method, along with k-nearest neighbor (kNN) and a support vector machine (SVM), to determine the smartphone addiction level. The risk group (13 subjects) and non-risk group (12 subjects) watched videos portraying the following concepts: Relaxed, Fear, Joy, and Sadness. We found that the risk group was more emotionally unstable than the non-risk group. In recognizing Fear, a clear difference appeared between the risk and non-risk group. The results showed that the gamma band was the most obviously different between the risk and non-risk groups. Moreover, we demonstrated that the measurements of activity in the frontal, parietal, and temporal lobes were indicators of emotion recognition. Through the DBN, we confirmed that these measurements were more accurate in the non-risk group than they were in the risk group. The risk group had higher accuracy in low valence and arousal; on the other hand, the non-risk group had higher accuracy in high valence and arousal.
Mobile phones are an essential part of an adolescent’s life, leading them to text, phone, or message into the night. Longitudinal latent growth models were used to examine relations between changes in adolescent night-time mobile phone use, changes in sleep behavior, and changes in well-being (depressed mood, externalizing behavior, self-esteem, and coping) for 1,101 students (43% male) between 13 and 16 years old. Both night-time mobile phone use and poor sleep behavior underwent positive linear growth over time. Increased night-time mobile phone use was directly associated with increased externalizing behavior and decreased self-esteem and coping. Changes in sleep behavior mediated the relation between early changes in night-time mobile phone use and later increases in depressed mood and externalizing behavior and later declines in self-esteem and coping. https://onlinelibrary.wiley.com/doi/full/10.1111/cdev.12836
Motor activity in physical and psychological stress exposure has been studied almost exclusively with self-assessment questionnaires and from reports that derive from human observer, such as verbal rating and simple descriptive scales. However, these methods are limited in objectively quantifying typical behaviour of stress. We propose to use accelerometer data from smartphones to objectively quantify stress levels. Used data was collected in real-world setting, from 29 employees in two different organisations over 5 weeks. To improve classification performance we propose to use intermediate models. These intermediate models represent the mood state of a person which is used to build the final stress prediction model. In particular, we obtained an accuracy of 78.2 % to classify stress levels.
Objective:
The aim of this systematic review is to synthesize research studies involving the use of smart watch devices for healthcare.
Materials and methods:
The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) was chosen as the systematic review methodology. We searched PubMed, CINAHL Plus, EMBASE, ACM, and IEEE Xplore. In order to include ongoing clinical trials, we also searched ClinicalTrials.gov. Two investigators evaluated the retrieved articles for inclusion. Discrepancies between investigators regarding article inclusion and extracted data were resolved through team discussion.
Results:
356 articles were screened and 24 were selected for review. The most common publication venue was in conference proceedings (13, 54%). The majority of studies were published or presented in 2015 (19, 79%). We identified two registered clinical trials underway. A large proportion of the identified studies focused on applications involving health monitoring for the elderly (6, 25%). Five studies focused on patients with Parkinson's disease and one on cardiac arrest. There were no studies which reported use of usability testing before implementation.
Discussion:
Most of the reviewed studies focused on the chronically ill elderly. There was a lack of detailed description of user-centered design or usability testing before implementation. Based on our review, the most commonly used platform in healthcare research was that of the Android Wear. The clinical application of smart watches as assistive devices deserves further attention.
Conclusion:
Smart watches are unobtrusive and easy to wear. While smart watch technology supplied with biosensors has potential to be useful in a variety of healthcare applications, rigorous research with their use in clinical settings is needed.
Machine learning addresses the question of how to build computers that improve automatically through experience. It is one
of today’s most rapidly growing technical fields, lying at the intersection of computer science and statistics, and at the
core of artificial intelligence and data science. Recent progress in machine learning has been driven both by the development
of new learning algorithms and theory and by the ongoing explosion in the availability of online data and low-cost computation.
The adoption of data-intensive machine-learning methods can be found throughout science, technology and commerce, leading
to more evidence-based decision-making across many walks of life, including health care, manufacturing, education, financial
modeling, policing, and marketing.
Artificial neural networks are appearing as useful alternatives to traditional statistical modelling techniques in many scientific disciplines. This paper presents a general introduction and discussion of recent applications of the multilayer perceptron, one type of artificial neural network, in the atmospheric sciences.
Although polysomnography is necessary for diagnosis of most sleep disorders, it is also expensive, time-consuming, intrusive, and interferes with sleep. Field-based activity monitoring is increasingly used as an alternative measure that can be used to answer certain clinical and research questions. The purpose of this study was to evaluate the reliability and validity of a novel activity monitoring device (Fitbit) compared to both polysomnography and standard actigraphy (Actiwatch-64).
To test validity, simultaneous Fitbit and actigraph were worn during standard overnight polysomnography by 24 healthy adults at the West Virginia University sleep research laboratory. To test inter-Fitbit reliability, three participants also wore two of the Fitbit devices overnight at home.
Fitbit showed high intradevice reliability = 96.5-99.1. Fitbit and actigraph differed significantly on recorded total sleep time and sleep efficiency between each other and polysomnography. Bland-Altman plots indicated that both Fitbit and actigraph overestimated sleep efficiency and total sleep time. Sensitivity of both Fitbit and actigraphy for accurately identifying sleep was high within all sleep stages and during arousals; specificity of both Fitbit and actigraph for accurately identifying wake was poor. Specificity of actigraph was higher except for wake before sleep onset; sensitivity of Fitbit was higher in all sleep stages and during arousals.
The web-based Fitbit, available at a markedly reduced price and with several convenience factors compared to standard actigraphy, may be an acceptable activity measurement instrument for use with normative populations. However, Fitbit has the same specificity limitations as actigraphy; both devices consistently misidentify wake as sleep and thus overestimate both sleep time and quality. Use of the Fitbit will also require specific validation before it can be used to assess disordered populations and or different age groups.
Support vector machines (SVMs) are becoming popular in a wide variety of biological applications. But, what exactly are SVMs and how do they work? And what are their most promising applications in the life sciences?