Predicting hypoxic hypoxia using machine learning and wearable sensors

Sleep apnea is a sleep disorder in which breathing is briefly and repeatedly interrupted. Polysomnography (PSG) is the standard clinical test for diagnosing sleep apnea. However, it is expensive and time-consuming which requires hospital visits, specialized wearable sensors, professional installations, and long waiting lists. To address this problem, we design a smartwatch-based system called ApneaDetector, which exploits the built-in sensors in smartwatches to detect sleep apnea. Through a clinical study, we identify features of sleep apnea captured by smartwatch, which can be leveraged by machine learning techniques for sleep apnea detection. However, there are many technical challenges such as how to extract various special patterns from the noisy and multi-axis sensing data. To address these challenges, we propose signal denoising and data calibration techniques to process the noisy data while preserving the peaks and troughs which reflect the possible apnea events. We identify the characteristics of sleep apnea such as signal spikes which can be captured by smartwatch, and propose methods to extract proper features to train machine learning models for apnea detection. Through extensive experimental evaluations, we demonstrate that our system can detect apnea events with high precision (0.9674), recall (0.9625), and F1-score (0.9649).

Download Full-text

Assessment of construction workers’ perceived risk using physiological data from wearable sensors: A machine learning approach

Journal of Building Engineering ◽

10.1016/j.jobe.2021.102824 ◽

2021 ◽

pp. 102824

Author(s):

By Gaang Leei ◽

Byungjoo Choi ◽

Houtan Jebelli ◽

SangHyun Lee

Keyword(s):

Machine Learning ◽

Perceived Risk ◽

Wearable Sensors ◽

Construction Workers ◽

Physiological Data ◽

Learning Approach ◽

Machine Learning Approach

Download Full-text

Recognition of human activities using machine learning methods with wearable sensors

2017 IEEE 7th Annual Computing and Communication Workshop and Conference (CCWC) ◽

10.1109/ccwc.2017.7868369 ◽

2017 ◽

Cited By ~ 15

Author(s):

Long Cheng ◽

Yani Guan ◽

Kecheng Zhu ◽

Yiyang Li

Keyword(s):

Machine Learning ◽

Human Activities ◽

Wearable Sensors ◽

Learning Methods ◽

Machine Learning Methods

Download Full-text

Machine Learning based Human Fall Detection System

International Journal for Research in Applied Science and Engineering Technology ◽

10.22214/ijraset.2021.35394 ◽

2021 ◽

Vol 9 (VI) ◽

pp. 2677-2682

Author(s):

Nishanth P

Keyword(s):

Machine Learning ◽

Detection System ◽

Wearable Sensors ◽

Fall Detection ◽

The Elderly ◽

World Health ◽

Support Vector ◽

Data Output ◽

Health Organization ◽

Context Features

Falls have become one of the reasons for death. It is common among the elderly. According to World Health Organization (WHO), 3 out of 10 living alone elderly people of age 65 and more tend to fall. This rate may get higher in the upcoming years. In recent years, the safety of elderly residents alone has received increased attention in a number of countries. The fall detection system based on the wearable sensors has made its debut in response to the early indicator of detecting the fall and the usage of the IoT technology, but it has some drawbacks, including high infiltration, low accuracy, poor reliability. This work describes a fall detection that does not reliant on wearable sensors and is related on machine learning and image analysing in Python. The camera's high-frequency pictures are sent to the network, which uses the Convolutional Neural Network technique to identify the main points of the human. The Support Vector Machine technique uses the data output from the feature extraction to classify the fall. Relatives will be notified via mobile message. Rather than modelling individual activities, we use both motion and context information to recognize activities in a scene. This is based on the notion that actions that are spatially and temporally connected rarely occur alone and might serve as background for one another. We propose a hierarchical representation of action segments and activities using a two-layer random field model. The model allows for the simultaneous integration of motion and a variety of context features at multiple levels, as well as the automatic learning of statistics that represent the patterns of the features.

Download Full-text

Human Activity Recognition of Children with Wearable Devices Using LightGBM Machine Learning

10.21203/rs.3.rs-1076792/v1 ◽

2021 ◽

Author(s):

Gábor Csizmadia ◽

Krisztina Liszkai-Peres ◽

Bence Ferdinandy ◽

Ádám Miklósi ◽

Veronika Konok

Keyword(s):

Machine Learning ◽

Activity Recognition ◽

Human Activity ◽

Learning Algorithm ◽

Binary Classification ◽

Wearable Sensors ◽

Window Size ◽

Human Activity Recognition ◽

Daily Routine ◽

Light Gradient

Abstract Human activity recognition (HAR) using machine learning (ML) methods is a relatively new method for collecting and analyzing large amounts of human behavioral data using special wearable sensors. Our main goal was to find a reliable method which could automatically detect various playful and daily routine activities in children. We defined 40 activities for ML recognition, and we collected activity motion data by means of wearable smartwatches with a special SensKid software. We analyzed the data of 34 children (19 girls, 15 boys; age range: 6.59 – 8.38; median age = 7.47). All children were typically developing first graders from three elementary schools. The activity recognition was a binary classification task which was evaluated with a Light Gradient Boosted Machine (LGBM)learning algorithm, a decision based method with a 3-fold cross validation. We used the sliding window technique during the signal processing, and we aimed at finding the best window size for the analysis of each behavior element to achieve the most effective settings. Seventeen activities out of 40 were successfully recognized with AUC values above 0.8. The window size had no significant effect. The overall accuracy was 0.95, which is at the top segment of the previously published similar HAR data. In summary, the LGBM is a very promising solution for HAR. In line with previous findings, our results provide a firm basis for a more precise and effective recognition system that can make human behavioral analysis faster and more objective.

Download Full-text

UP-Fall Detection Dataset: A Multimodal Approach

Sensors ◽

10.3390/s19091988 ◽

2019 ◽

Vol 19 (9) ◽

pp. 1988 ◽

Cited By ~ 33

Author(s):

Lourdes Martínez-Villaseñor ◽

Hiram Ponce ◽

Jorge Brieva ◽

Ernesto Moya-Albor ◽

José Núñez-Martínez ◽

...

Keyword(s):

Machine Learning ◽

Learning Community ◽

Wearable Sensors ◽

Fall Detection ◽

Machine Learning Techniques ◽

Elderly Persons ◽

Negative Consequences ◽

Detection Systems ◽

Important Health ◽

Experimental Use

Falls, especially in elderly persons, are an important health problem worldwide. Reliable fall detection systems can mitigate negative consequences of falls. Among the important challenges and issues reported in literature is the difficulty of fair comparison between fall detection systems and machine learning techniques for detection. In this paper, we present UP-Fall Detection Dataset. The dataset comprises raw and feature sets retrieved from 17 healthy young individuals without any impairment that performed 11 activities and falls, with three attempts each. The dataset also summarizes more than 850 GB of information from wearable sensors, ambient sensors and vision devices. Two experimental use cases were shown. The aim of our dataset is to help human activity recognition and machine learning research communities to fairly compare their fall detection solutions. It also provides many experimental possibilities for the signal recognition, vision, and machine learning community.

Download Full-text

Can Machine Learning Predict Stress Reduction Based on Wearable Sensors’ Data Following Relaxation at Workplace? A Pilot Study

Processes ◽

10.3390/pr8040448 ◽

2020 ◽

Vol 8 (4) ◽

pp. 448

Author(s):

Alessandro Tonacci ◽

Alessandro Dellabate ◽

Andrea Dieni ◽

Lorenzo Bachi ◽

Francesco Sansone ◽

...

Keyword(s):

Machine Learning ◽

Stress Reduction ◽

Wearable Sensors ◽

Well Being ◽

Significant Heterogeneity ◽

Learning Approaches ◽

Significant Difference ◽

Group A ◽

Group B ◽

Audio Video

Nowadays, psychological stress represents a burdensome condition affecting an increasing number of subjects, in turn putting into practice several strategies to cope with this issue, including the administration of relaxation protocols, often performed in non-structured environments, like workplaces, and constrained within short times. Here, we performed a quick relaxation protocol based on a short audio and video, and analyzed physiological signals related to the autonomic nervous system (ANS) activity, including electrocardiogram (ECG) and galvanic skin response (GSR). Based on the features extracted, machine learning was applied to discriminate between subjects benefitting from the protocol and those with negative or no effects. Twenty-four healthy volunteers were enrolled for the protocol, equally and randomly divided into Group A, performing an audio-video + video-only relaxation, and Group B, performing an audio-video + audio-only protocol. From the ANS point of view, Group A subjects displayed a significant difference in the heart rate variability-related parameter SDNN across the test phases, whereas both groups displayed a different GSR response, albeit at different levels, with Group A displaying greater differences across phases with respect to Group B. Overall, the majority of the volunteers enrolled self-reported an improvement of their well-being status, according to structured questionnaires. The use of neural networks helped in discriminating those with a positive effect of the relaxation protocol from those with a negative/neutral impact based on basal autonomic features with a 79.2% accuracy. The results obtained demonstrated a significant heterogeneity in autonomic effects of the relaxation, highlighting the importance of maintaining a structured, well-defined protocol to produce significant benefits at the ANS level. Machine learning approaches can be useful to predict the outcome of such protocols, therefore providing subjects less prone to positive responses with personalized advice that could improve the effect of such protocols on self-relaxation perception.

Download Full-text