On-Body Sensor Positions Hierarchical Classification

Abstract

Many motion sensor-based applications have been developed in recent years because they provide useful information about daily activities and current health status of users. However, most of these applications require knowledge of sensor positions. Therefore, this research focused on the problem of detecting sensor positions. We collected standing-still and walking sensor data at various body positions from ten subjects. The offset values were removed by subtracting the sensor data of standing-still phase from the walking data for each axis of each sensor unit. Our hierarchical classification technique is based on optimizing local classifiers. Many common features are computed, and informative features are selected for specific classifications. In this approach, local classifiers such as arm-side and hand-side discriminations yielded F1-scores of 0.99 and 1.00, correspondingly. Overall, the proposed method achieved an F1-score of 0.81 and 0.84 using accelerometers and gyroscopes, respectively. Furthermore, we also discuss contributive features and parameter tuning in this analysis.

An Intelligent Clinical Decision Support System for Predicting Acute Graft-versus-host Disease (aGvHD) following Allogeneic Hematopoietic Stem Cell Transplantation

“…By appending classifiers on top of each other iteratively, the next classifier can modify the errors of the previous one. This process is recurred until the training data set is accurately predicted [ 23 ].…”

A Human Activity Recognition Algorithm Based on Stacking Denoising Autoencoder and LightGBM

“…The second factor is the collection frequency. The researchers have tried multiple possibilities such as 50 Hz [ 17 ], 76.25 Hz [ 18 ], 100 Hz [ 19 ], and 120 Hz [ 20 ]. Besides, there are various publicly available datasets that provide strong support for human activities recognition such as UCI [ 16 ], WISDM [ 21 ], HASC [ 22 ], and RealWorld HAR [ 23 ].…”

“…According to the feature construct method, the current feature extraction methods of HAR are divided into the artificial feature construction method [ 15 , 20 , 21 , 25 , 26 ] and deep learning feature construction method [ 13 , 17 , 27 , 28 , 29 ]. Tran et al [ 26 ] added the features of the frequency domain into consideration and Sang et al [ 20 ] imported the fractal dimension. Khalifa et al [ 30 ] introduced the concept of kinetic energy collection.…”

“…As traditional machine learning methods, Support Vector Machine (SVM) [ 25 , 26 ], Random Forest (RF) [ 37 ], Logistic Regression (LR) [ 20 ], eXtreme Gradient Boost (XGB) [ 20 ], and Light Gradient Boosting Machine (LGB) have been deeply studied. Sang et al [ 20 ] compared the classification accuracy of different classifiers with the same features, such as LR, Decision Tree (DT), SVM, and XGB. Bayat et al came from the perspective of classifier fusion and found that Multilayer Perceptron (MP) + LigitBoost + SVM will achieve the best classification effect.…”

Human Activities and Postures Recognition: From Inertial Measurements to Quaternion-Based Approaches

“…Moreover, focusing on statistical features only can over shade the physical significance of data and thus provide us with lower detection accuracy rates. As a matter of fact, numerous studies have been done on this matter, in [17], it has been shown that 112 features, extracted from the accelerometer and the gyroscope, are considered important; however, for specific applications they can be lowered to 19 features for the accelerometer and 23 for the gyroscope. For arm and hand side classifications, accelerometer features can be reduced to 4 or even 1.…”