Comparison Of Machine Learning Algorithms For Heart Rate Variability Based Driver Drowsiness Detection

Introduction: The Pulse Initiative on resuscitation identified the need to develop biosensing for detection of critical limitations of blood flow. The ability to rapidly detect the severity of hemorrhage based on heart rate has been limited. Use of heart rate variability (HRV) is problematic. We used a number of new defined ECG features based on discrete wavelet transformation (DWT) that may be used to estimate blood loss severity. The features are defined based on the energy of detail coefficients of Daubecies DWT. Methods: The performance of DWT was tested using ECG data from a human model of hemorrhage using lower body negative pressure (LBNP). LBNP consisted of a 5-minute rest period (0 mm Hg) followed by 5 minutes of chamber decompression of the lower body to −15, −30, −45, and −60 mm Hg and additional increments of −10 mm Hg every 5 minutes until the onset of cardiovascular collapse. These levels were divided into 3 classes (mild: −15 to −30 mmHg; moderate: −45 to −60 mmHg; severe: over −60 mmHg). These levels correspond to estimated blood losses of 400 –550 cc, 500 –1000 cc and greater than 1000 cc respectively. The ECG DWT features of subjects were used for classification of each ECG recording during volume loss levels. Before classification in order to eliminate redundancy among the features, principal component analysis is applied to the feature set. Machine learning algorithms (SVM, AdaBoost, C4.5) were then applied to analyze the processed features and predict the severity of blood loss. Results: A 219 sample set was used to classify groups by using machine learning algorithms with 10-fold cross validation. C4.5 outperformed other algorithms with a prediction accuracy of 74.4%. The average precision and recall (sensitivity) for the three classes were 77.4% and 76.1%, respectively. In particular, 30 out of 39 cases in the severe class were correctly classified by C4.5. These results required sampling rates of only 125 Hz. Conclusion: This is the first reported use of an ECG analysis method to classify volume loss. The DWT method described may have the ability to rapidly determine the degree of volume loss from hemorrhage providing for more rapid triage and decision making. This may be particularly helpful for remote monitoring of war fighters or for mass casualty care.

Download Full-text

Driver Drowsiness Detection System using Machine Learning Algorithms

International Journal of Recent Technology and Engineering - 2 ◽

10.35940/ijrte.f7514.038620 ◽

2020 ◽

Vol 8 (6) ◽

pp. 990-993

Keyword(s):

Machine Learning ◽

Real Time ◽

Detection System ◽

Learning Algorithms ◽

Machine Learning Algorithms ◽

World Health ◽

Raspberry Pi ◽

Data Set ◽

Drowsiness Detection ◽

Driver Drowsiness

Road crashes are the most common forms of accidents and deaths worldwide, and the significant reasons for these accidents are usually drunken, drowsiness and reckless behaviour of the driver. According to the World Health Organization, road traffic injuries have risen to 1.25 billion worldwide, which makes driver drowsiness detection a major potential area to avert numerous sleep-induced road accidents. This project proposes an idea to detect drowsiness using machine learning algorithms, hence alarming the driver in real-time to prevent a collision. The model uses the Haar Cascade algorithm, along with the OpenCV library to monitor the real-time video of the driver and to detect the eyes of the driver. The system uses the Eye Aspect Ratio (EAR) concept to determine if the eyes are open or closed. We also feed a data-set file consisting of the facial features data-points to train the machine learning algorithm. The model inspects each frame of the video, which helps to recognize the state of the driver. Furthermore, a Raspberry Pi single-board computer, combined with a camera module and an alarm system, facilitates the project to emulate a compact drowsiness detection system suitable for different automobiles.

Download Full-text

Assessment of the Potential of Wrist-Worn Wearable Sensors for Driver Drowsiness Detection

Sensors ◽

10.3390/s20041029 ◽

2020 ◽

Vol 20 (4) ◽

pp. 1029 ◽

Cited By ~ 1

Author(s):

Thomas Kundinger ◽

Nikoletta Sofra ◽

Andreas Riener

Keyword(s):

Machine Learning ◽

User Study ◽

Driving Simulator ◽

Machine Learning Algorithms ◽

Sensor Technology ◽

Physiological Data ◽

Promising Alternative ◽

Drowsiness Detection ◽

Driver Drowsiness ◽

Medical Grade

Drowsy driving imposes a high safety risk. Current systems often use driving behavior parameters for driver drowsiness detection. The continuous driving automation reduces the availability of these parameters, therefore reducing the scope of such methods. Especially, techniques that include physiological measurements seem to be a promising alternative. However, in a dynamic environment such as driving, only non- or minimal intrusive methods are accepted, and vibrations from the roadbed could lead to degraded sensor technology. This work contributes to driver drowsiness detection with a machine learning approach applied solely to physiological data collected from a non-intrusive retrofittable system in the form of a wrist-worn wearable sensor. To check accuracy and feasibility, results are compared with reference data from a medical-grade ECG device. A user study with 30 participants in a high-fidelity driving simulator was conducted. Several machine learning algorithms for binary classification were applied in user-dependent and independent tests. Results provide evidence that the non-intrusive setting achieves a similar accuracy as compared to the medical-grade device, and high accuracies (>92%) could be achieved, especially in a user-dependent scenario. The proposed approach offers new possibilities for human–machine interaction in a car and especially for driver state monitoring in the field of automated driving.

Download Full-text

Classification of caesarean section and normal vaginal deliveries using foetal heart rate signals and advanced machine learning algorithms

BioMedical Engineering OnLine ◽

10.1186/s12938-017-0378-z ◽

2017 ◽

Vol 16 (1) ◽

Cited By ~ 14

Author(s):

Paul Fergus ◽

Abir Hussain ◽

Dhiya Al-Jumeily ◽

De-Shuang Huang ◽

Nizar Bouguila

Keyword(s):

Machine Learning ◽

Heart Rate ◽

Caesarean Section ◽

Learning Algorithms ◽

Machine Learning Algorithms ◽

Foetal Heart Rate ◽

Foetal Heart ◽

Vaginal Deliveries

Download Full-text

The effect of principal component analysis in the diagnosis of congestive heart failure via heart rate variability analysis

Proceedings of the Institution of Mechanical Engineers Part H Journal of Engineering in Medicine ◽

10.1177/09544119211036806 ◽

2021 ◽

pp. 095441192110368

Author(s):

Mustafa B Selek ◽

Bartu Yesilkaya ◽

Saadet S Egeli ◽

Yalcin Isler

Keyword(s):

Machine Learning ◽

Heart Failure ◽

Heart Rate ◽

Principal Component Analysis ◽

Congestive Heart Failure ◽

Heart Rate Variability ◽

Random Forest ◽

Principal Components ◽

Principal Component ◽

Machine Learning Algorithms

In this study, we investigated the effect of principal component analysis (PCA) in congestive heart failure (CHF) diagnosis using various machine learning algorithms from 5-min HRV data. The extracted 59 heart rate variability (HRV) features consist of statistical time-domain measures, frequency-domain measures (power spectral density estimations from Fourier transform and Lomb-Scargle methods), time-frequency HRV measures (Wavelet transform), and nonlinear HRV measures (Poincare plot, symbolic dynamics, detrended fluctuation analysis, and sample entropy). All these HRV features are the classifiers’ inputs. We repeated the study ten times using the first one to the first 10 principal components from PCA instead of all HRV features. Nine different classifiers, namely logistic regression, Naive Bayes, k-nearest neighbors, decision tree, AdaBoost, support vector machines, stochastic gradient descent, random forest, and artificial neuronal network (multilayer perceptron) are examined. The proposed study results in the 100% accuracy, 100% specificity, and 100% sensitivity after utilizing PCA (with the first eight principal components) using the Random Forest classifier where the maximum classifier performances are the 86% accuracy, 79% specificity, and 86% sensitivity before PCA. In conclusion, PCA is beneficial in the diagnosis of patients with CHF. In addition, we experienced the online Python-based visual machine learning tool, Orange, which can implement well-known machine learning algorithms.

Download Full-text