On Mental Health Monitoring Using Commercial Wearable Devices and Machine Intelligence
-
Partha Pratim Sarmah
Abstract
Mental health problems such as anxiety, stress, and depression are increasingly prevalent nowadays due to fast-paced lifestyles, work pressure, social isolation, and other factors. Technological advancements in wearable devices, including smartwatches, smart bands, and smart rings, offer a promising approach to continuous physiological monitoring. Since the cohesion between mental/emotional health and physiological parameters is already an established fact through various researches, wearable devices should necessarily enable the early detection of mental health conditions as well. Our research studies the feasibility of commercially available smartwatches for emotion detection. Commercial wearable devices do not consider mental/emotional health as their primary concern in their design goal. To study its feasibility in such devices, in our experiment, we use the LifeSnaps dataset, a real-life multimodal dataset that captures physiological and psychological data related to 71 users from various demographic locations. First, we examine several physiological features such as heart rate, oxygen levels, calories burned, and sleep patterns (including light sleep, deep sleep, etc.) in the time domain, as well as some psychological surveys such as PANAS and STAI and emotional state survey collected using SEMA3 mobile app, and try to establish their cohesion using several machine learning models. We achieve a maximum accuracy of 78.75% using random forest. Furthermore, to improve accuracy, we extract additional signal-level features using fast Fourier transformation and apply recursive feature elimination to identify the most relevant features through our proposed methods: (I) selective feature-based approach and (II) holistic feature-based approach. Next, with these enhancements, the machine learning model achieves the highest accuracy of 81.01% using XGBoost. Additionally, we explore two neural network models (bidirectional long short-term memory and FNN (feedforward neural network)) with random oversampling and achieve 97% accuracy with FNN.
Abstract
Mental health problems such as anxiety, stress, and depression are increasingly prevalent nowadays due to fast-paced lifestyles, work pressure, social isolation, and other factors. Technological advancements in wearable devices, including smartwatches, smart bands, and smart rings, offer a promising approach to continuous physiological monitoring. Since the cohesion between mental/emotional health and physiological parameters is already an established fact through various researches, wearable devices should necessarily enable the early detection of mental health conditions as well. Our research studies the feasibility of commercially available smartwatches for emotion detection. Commercial wearable devices do not consider mental/emotional health as their primary concern in their design goal. To study its feasibility in such devices, in our experiment, we use the LifeSnaps dataset, a real-life multimodal dataset that captures physiological and psychological data related to 71 users from various demographic locations. First, we examine several physiological features such as heart rate, oxygen levels, calories burned, and sleep patterns (including light sleep, deep sleep, etc.) in the time domain, as well as some psychological surveys such as PANAS and STAI and emotional state survey collected using SEMA3 mobile app, and try to establish their cohesion using several machine learning models. We achieve a maximum accuracy of 78.75% using random forest. Furthermore, to improve accuracy, we extract additional signal-level features using fast Fourier transformation and apply recursive feature elimination to identify the most relevant features through our proposed methods: (I) selective feature-based approach and (II) holistic feature-based approach. Next, with these enhancements, the machine learning model achieves the highest accuracy of 81.01% using XGBoost. Additionally, we explore two neural network models (bidirectional long short-term memory and FNN (feedforward neural network)) with random oversampling and achieve 97% accuracy with FNN.
Chapters in this book
- Frontmatter I
- Contents V
- Early Prediction of Chronic Kidney Disease Using a Novel Hybrid Regularized Adaptive Boosting Algorithm: An Advanced Machine Learning Approach 1
- DigiCure: A Patient-Centric Framework for Digital Transformation in Healthcare 21
- Exploring Machine Learning Approaches for Maximizing the Likelihood of Diabetes Classification 41
- A Hybrid Machine Learning Model for Risk Stratification and Functional Outcome Prediction in Stroke Survivors 61
- Data-Driven Machine Learning Strategies for Oncological Disease Prediction and Early-Stage Detection 83
- Machine Learning Applications in Mental Health: Ensemble-Based Predictive Modeling for Depression and Anxiety detection 103
- Privacy-Preserving Machine Learning in Clinical Research: Using Federated Learning to Protect Patient Data 129
- EpiCastNet: A Spatiotemporal Hybrid Learning Framework for Real-Time Epidemic Forecasting 149
- Machine Learning for Early Detection of Chronic Diseases: A Case Study in Diabetes Prediction 171
- Machine Learning Techniques for Healthcare 193
- Applications and Benefits of Machine Learning in Healthcare 215
- Intelligent Treatment Recommendation Using CareRecNet: A Patient-Centered Approach to Digital Health Transformation 233
- Reinforcement-Driven Graph Neural Framework for Personalized and Proactive Patient Care in Digital Health Systems 251
- Hybrid Attention-Driven Network for Predictive Healthcare Using Machine Learning and Data Analytics Perspective 271
- MSAG-DFE: A Multi-scale Attention-Guided Deep Feature Extraction Framework for Enhanced Medical Image Diagnostics 287
- On Mental Health Monitoring Using Commercial Wearable Devices and Machine Intelligence 305
- Enhancing Healthcare Delivery Through Evidence-Based Data Utilization 335
- AGBO-CP: An Adaptive Gradient Boosted Optimization Framework for Enhanced Clinical Prediction Accuracy 367
- A Hierarchical Cross-Fusion Feature Extraction Network for Accurate Cervical Cancer Classification Using Cytology Images 387
- Analyzing the Impact of Social Network on Epidemiological Spread in the Healthcare Sector 409
- Intelligent Interventions: Practical Applications of Machine Learning for Data-Driven Decision-Making in Healthcare 431
- Stress Recognition Through Physiological and Behavioral Signals: A Machine Learning Perspective 453
- MediChain-FL: A Federated Blockchain Framework for Privacy-Preserving and Intelligent Healthcare Data Exchange 485
- Reinforced Multi-objective Optimization Framework for Adaptive Healthcare Decision Intelligence 503
- Index
Chapters in this book
- Frontmatter I
- Contents V
- Early Prediction of Chronic Kidney Disease Using a Novel Hybrid Regularized Adaptive Boosting Algorithm: An Advanced Machine Learning Approach 1
- DigiCure: A Patient-Centric Framework for Digital Transformation in Healthcare 21
- Exploring Machine Learning Approaches for Maximizing the Likelihood of Diabetes Classification 41
- A Hybrid Machine Learning Model for Risk Stratification and Functional Outcome Prediction in Stroke Survivors 61
- Data-Driven Machine Learning Strategies for Oncological Disease Prediction and Early-Stage Detection 83
- Machine Learning Applications in Mental Health: Ensemble-Based Predictive Modeling for Depression and Anxiety detection 103
- Privacy-Preserving Machine Learning in Clinical Research: Using Federated Learning to Protect Patient Data 129
- EpiCastNet: A Spatiotemporal Hybrid Learning Framework for Real-Time Epidemic Forecasting 149
- Machine Learning for Early Detection of Chronic Diseases: A Case Study in Diabetes Prediction 171
- Machine Learning Techniques for Healthcare 193
- Applications and Benefits of Machine Learning in Healthcare 215
- Intelligent Treatment Recommendation Using CareRecNet: A Patient-Centered Approach to Digital Health Transformation 233
- Reinforcement-Driven Graph Neural Framework for Personalized and Proactive Patient Care in Digital Health Systems 251
- Hybrid Attention-Driven Network for Predictive Healthcare Using Machine Learning and Data Analytics Perspective 271
- MSAG-DFE: A Multi-scale Attention-Guided Deep Feature Extraction Framework for Enhanced Medical Image Diagnostics 287
- On Mental Health Monitoring Using Commercial Wearable Devices and Machine Intelligence 305
- Enhancing Healthcare Delivery Through Evidence-Based Data Utilization 335
- AGBO-CP: An Adaptive Gradient Boosted Optimization Framework for Enhanced Clinical Prediction Accuracy 367
- A Hierarchical Cross-Fusion Feature Extraction Network for Accurate Cervical Cancer Classification Using Cytology Images 387
- Analyzing the Impact of Social Network on Epidemiological Spread in the Healthcare Sector 409
- Intelligent Interventions: Practical Applications of Machine Learning for Data-Driven Decision-Making in Healthcare 431
- Stress Recognition Through Physiological and Behavioral Signals: A Machine Learning Perspective 453
- MediChain-FL: A Federated Blockchain Framework for Privacy-Preserving and Intelligent Healthcare Data Exchange 485
- Reinforced Multi-objective Optimization Framework for Adaptive Healthcare Decision Intelligence 503
- Index
