Accelerometer Signals Research Articles

Multilevel attention mechanism for motion fatigue recognition based on sEMG and ACC signal fusion.

This study aims to develop a cost-effective and reliable motion monitoring device capable of comprehensive fatigue analysis. It achieves this objective by integrating surface electromyography (sEMG) and accelerometer (ACC) signals through a feature fusion strategy. The study introduces a multi-level attention mechanism for classification, leveraging convolutional neural networks (CNNs). The preprocessing phase involves a local feature attention mechanism that enhances local waveform features using the amplitude envelope. A dual-scale attention mechanism, operating at both channel and neuron levels, is employed to enhance the model's learning from high-dimensional fused data, improving feature extraction and generalization. The local feature attention mechanism significantly improves the model's classification accuracy and convergence, as demonstrated in ablation experiments. The model, optimized with multi-level attention mechanisms, excels in accuracy and generalization, particularly in handling data with pseudo-artifacts. Computational analysis indicates that the proposed optimization algorithm has minimal impact on CNN's training and testing times. The study achieves recognition accuracies of 92.52%, 92.38%, and 92.30%, as well as F1-scores of 91.92%, 92.13%, and 92.29% for the three fatigue states, affirming its reliability. This research provides technical support for the development of affordable and dependable wearable motion monitoring devices.

Quantifying Asymmetric Gait Pattern Changes Using a Hidden Markov Model Similarity Measure (HMM-SM) on Inertial Sensor Signals.

Wearable gait analysis systems using inertial sensors offer the potential for easy-to-use gait assessment in lab and free-living environments. This can enable objective long-term monitoring and decision making for individuals with gait disabilities. This study explores a novel approach that applies a hidden Markov model-based similarity measure (HMM-SM) to assess changes in gait patterns based on the gyroscope and accelerometer signals from just one or two inertial sensors. Eleven able-bodied individuals were equipped with a system which perturbed gait patterns by manipulating stance-time symmetry. Inertial sensor data were collected from various locations on the lower body to train hidden Markov models. The HMM-SM was evaluated to determine whether it corresponded to changes in gait as individuals deviated from their baseline, and whether it could provide a reliable measure of gait similarity. The HMM-SM showed consistent changes in accordance with stance-time symmetry in the following sensor configurations: pelvis, combined upper leg signals, and combined lower leg signals. Additionally, the HMM-SM demonstrated good reliability for the combined upper leg signals (ICC = 0.803) and lower leg signals (ICC = 0.795). These findings provide preliminary evidence that the HMM-SM could be useful in assessing changes in overall gait patterns. This could enable the development of compact, wearable systems for unsupervised gait assessment, without the requirement to pre-identify and measure a set of gait parameters.

Fall Detection Based on Data-Adaptive Gaussian Average Filtering Decomposition and Machine Learning

Falls are a significant health concern leading to increased morbidity and healthcare costs, especially for the elderly. Early and accurate detection of fall events is critical for timely intervention and preventing severe complications. This study presents a novel approach to triaxial accelerometer signals by employing data-adaptive Gaussian average filtering (DAGAF) decomposition in conjunction with machine learning techniques for fall detection. The triaxial accelerometer signals from the FallAllD dataset were decomposed into intrinsic mode functions (IMFs) and a residual component, from which feature vectors were extracted to train support vector machine (SVM) and k-nearest neighbor (kNN) classifiers. Experimental results demonstrate that the combination of the first and the third IMFs with the residual component yields the highest classification accuracy of 96.34%, with SVM outperforming kNN across all performance metrics. This approach significantly improves fall detection accuracy compared to using raw accelerometer signals, highlighting its potential in enhancing wearable fall detection systems. The proposed DAGAF decomposition method not only enhances feature extraction but also provides a promising advancement in the field, suggesting its potential to increase the reliability and accuracy of fall detection in practical applications.

Robustness evaluation of acceleration-based early rub detection methodologies with real fluid-induced noise

The early detection of rotor-casing rub allows to prevent permanent damage and even catastrophic failure of rotating machines, saving maintenance costs and decreasing downtime. Very often, aeroderivative gas turbines can be monitored only with accelerometers on the casing, which provide signals that contain both stochastic and deterministic noise that masks the vibration measurements. Although real vibration data from turbomachines present a background noise that is neither white nor Gaussian, the rule of thumb is to evaluate the robustness of fault detection under white signal noise, which may lead to estimation errors of such robustness. Here we evaluate and compare the performance of several signal processing methodologies for very early rub detection on accelerometer signals under fluid-induced signal noise due to turbulent fluid excitation. The independence between the fluid-induced vibrations and the rub and unbalance vibration was proved, and experimental data of fluid-induced noise were added as signal noise to clean rub signals. The robustness and sensitivity of rub detection under fluid-induced noise were calculated and compared. This study proves the superiority of reassigned time–frequency analysis methods such as the Wavelet Synchrosqueezed Transform for very early rub detection as compared with traditional time–frequency analysis methods and with the Fourier Transform. To the authors’ knowledge, this is the first time that non-Gaussian fluid-induced signal noise is used to evaluate fault detection performance.

The Determination of On-Water Rowing Stroke Kinematics Using an Undecimated Wavelet Transform of a Rowing Hull-Mounted Accelerometer Signal.

Boat acceleration profiles can provide valuable information for coaches and practitioners to make meaningful technical interventions and monitor the determinants of success in rowing. Previous studies have used simple feature detection methods to identify key phases within individual strokes, such as drive onset, drive time, drive offset and stroke time. However, based on skill level, technique or boat class, the hull acceleration profile can differ, making robust feature detection more challenging. The current study's purpose is to employ the undecimated wavelet transform (UWT) technique to detect individual features in the stroke acceleration profile from a single rowing hull-mounted accelerometer. In this investigation, the temporal and kinematic values obtained using the AdMosTM sensor in conjunction with the UWT processing approach were strongly correlated with the comparative measures of the Peach™ instrumented oarlock system. The measures for stroke time displayed very strong agreeability between the systems for all boat classes, with ICC values of 0.993, 0.963 and 0.954 for the W8+, W4- and W1x boats, respectively. Similarly, the drive time was also very consistent, with strong to very strong agreeability, producing ICC values of 0.937, 0.901 and 0.881 for the W8+, W4- and W1x boat classes. Further, a Bland-Altman analysis displayed little to no bias between the AdMosTM-derived and Peach™ measures, indicating that there were no systematic discrepancies between signals. This single-sensor solution could form the basis for a simple, cost-effective and accessible alternative to multi-sensor instrumented systems for the determination of sub-stroke kinematic phases.

Control of combustion phasing using accelerometer-based non-intrusive sensing

Abstract Measuring the combustion phasing of an engine using in-cylinder pressure sensors is well established. However, pressure sensors need to be directly exposed to the in-cylinder environment, requiring changes to the cylinder head. Several methods have been proposed for sensing combustion phasing non-intrusively by mounting an accelerometer on the engine block. This paper presents real-time control of combustion phasing in a compression-ignition engine using non-intrusive accelerometer-based sensing during a dynamic fuel switch. A systematic data-driven control framework capable of handling fuel switching in real time is used. The control is designed based on the pressure data, and the real-time implementation is performed using the accelerometer signal. Results from combustion phasing tracking experiments performed on a compression-ignition engine are presented.

Speech synthesis from three-axis accelerometer signals using conformer-based deep neural network

Silent speech interfaces (SSIs) have emerged as innovative non-acoustic communication methods, and our previous study demonstrated the significant potential of three-axis accelerometer-based SSIs to identify silently spoken words with high classification accuracy. The developed accelerometer-based SSI with only four accelerometers and a small training dataset outperformed a conventional surface electromyography (sEMG)-based SSI. In this study, motivated by the promising initial results, we investigated the feasibility of synthesizing spoken speech from three-axis accelerometer signals. This exploration aimed to assess the potential of accelerometer-based SSIs for practical silent communication applications. Nineteen healthy individuals participated in our experiments. Five accelerometers were attached to the face to acquire speech-related facial movements while the participants read 270 Korean sentences aloud. For the speech synthesis, we used a convolution-augmented Transformer (Conformer)-based deep neural network model to convert the accelerometer signals into a Mel spectrogram, from which an audio waveform was synthesized using HiFi-GAN. To evaluate the quality of the generated Mel spectrograms, ten-fold cross-validation was performed, and the Mel cepstral distortion (MCD) was chosen as the evaluation metric. As a result, an average MCD of 5.03 ± 0.65 was achieved using four optimized accelerometers based on our previous study. Furthermore, the quality of generated Mel spectrograms was significantly enhanced by adding one more accelerometer attached under the chin, achieving an average MCD of 4.86 ± 0.65 (p < 0.001, Wilcoxon signed-rank test). Although an objective comparison is difficult, these results surpass those obtained using conventional SSIs based on sEMG, electromagnetic articulography, and electropalatography with the fewest sensors and a similar or smaller number of sentences to train the model. Our proposed approach will contribute to the widespread adoption of accelerometer-based SSIs, leveraging the advantages of accelerometers like low power consumption, invulnerability to physiological artifacts, and high portability.

Diesel Engine Turbocharger Monitoring by Processing Accelerometric Signals through Empirical Mode Decomposition and Independent Component Analysis

In this study, a method for the monitoring of internal combustion engine operation by vibration signals is proposed. The work falls within the context of the increasingly stringent standards relating to the environmental impact of engines and the development of monitoring and control techniques to ensure increased engine performance as well as fuel saving and reduction of pollutant emissions. Experimentation was performed on a turbocharged light-duty compression ignition direct-injection engine. Two monoaxial accelerometers were installed on the engine compressor case, the speed of which has been demonstrated to be closely related to the engine operation. Vibration measurements of the engine compressor case have been processed by combining the Empirical Mode Decomposition technique with Independent Component Analysis and Short Time Fourier Transform to indirectly estimate the turbocharger speed. The obtained traces have been compared to the direct turbocharger velocity measures during the stationary running of the engine (speed and load conditions varied in the complete engine’s range of operation). The results point out the potentiality of the methodology in algorithms devoted to identifying modifications of the combustion development regarding regular operation via indirect turbocharger speed monitoring.

Development of gravitationally aligned pendant IMU frailty identifier

This study focused on the practical development of a device for the early detection of frailty by leveraging the advanced Inertial Measurement Unit (IMU) technique. A five-time sit-to-stand test (FTSST) was conducted with people simply wearing a devised IMU pendant to evaluate the degree of frailty. As anticipated, the pendant sensor was recommended to be loosely strapped on older people's bodies for a comfortable outfit. Hence, the measurements from the IMU device would vary with the strap angle of the pedant. A calibration of gravitational alignment, which aligns the measurements to the gravity direction exactly with the FTSST motion signals consistently with an identical principal axis, would benefit the straightforward successive signal analysis and assessment. A prototype was thus developed with the designated calibration by using an open and systematic principle. This study analyzed the signals of accelerometer and gyroscope from IMU, and the subject's demographic profiles comprised a spectrum of features for identification analysis. Pearson correlation and principal component analysis for relevance ranking were used to reduce the feature dimensionality. Classifiers, including Support Vector Machine (SVM), Random Forest (RF), and Artificial Neural Network (ANN), were introduced to infer the frailty level of the subject. An innovative signal calibration and alignment process with solid mathematical support was imposed in the scheme. As deployed, the designated frailty identifier arrived at its generalized accuracy of 95.45 % with real clinic subjects in prediction. The alignment calibration process is the key for not only resolving the loose-attachment problem tolerantly but also enhancing the discriminability of the features. The prototype revealed the better integrity in developing a cheap, tiny, wearable device to identify the frailty.

Transfer learning with cross-domain adaptation of vibro-acoustic signals for electric motor mechanical fault detection and diagnosis.

Contact-based vibration measurement techniques involving accelerometers are typically utilized for machine health monitoring and manufacturing end-of-line quality control. Acoustic signals, encompassing sound pressure and particle velocity, carry crucial information about the operational status of underlying equipment and can be utilized for detecting various machine faults. Utilizing one-dimensional Convolution Neural Networks (1D-CNN) based on deep learning for processing raw acoustic time signals has proven to be a valuable approach in detecting machinery faults, offering an effective alternative to using traditional accelerometer signals. However, the success of deep learning methods is heavily dependent on the quantity and availability of data for robust network training. When transitioning to new non-contact type sensor technology, manufacturers may face challenges due to insufficient data for training deep learning models. In such cases, leveraging existing historical accelerometer-based data becomes crucial. Transfer learning emerges as a promising solution, allowing deep learning models trained on a source domain (SD) to be applied to a separate but related target domain (TD). This paper hence introduces a domain adaptation methodology by implementing transfer learning, where the SD 1D-CNN network is trained on accelerometer signals, then the trained model weights are transferred to the TD 1D-CNN network that can process acoustic signals.

Cutout effects on the vibration of sandwich auxetic cylindrical shells with an experimental validation

This study investigates the influence of auxetic core layers on the vibrational characteristics of sandwich cylindrical shells with cut-outs. The analysis utilizes the First-order Shear Deformation Theory (FSDT) and analytical techniques. The irregular behavior of auxetic re-entrante honeycomb cells necessitates the exploration of their effects on various systems. Additionally, the proposed method offers advantages over Finite Element Method (FEM). FEM modeling suffers from increased computational time and reduced accuracy with a large number of elements or high aspect ratio elements. These challenges are particularly significant for shells with geometric features of disparate scales, such as large cylinders with tiny cutouts or cutouts with low aspect ratios. However, the present research addresses these challenges by employing a method that integrates five panels with varying dimensions. Consequently, each section can utilize specific Two-Dimensional Generalized Differential Quadrature (2D-GDQ) grid points, tailored to the size of each side, enabling rapid and accurate prediction of displacement and stress parameters, particularly for shorter sides. The proposed method excels at precisely modeling cutouts while preserving their realistic properties. This is achieved by avoiding geometric and material simplifications and by assigning distinct boundary conditions to each region, including critical areas like corners, internal, and external sides. Moreover, Frequency Response Functions (FRFs) obtained from accelerometer and laser signals, alongside vibration-induced sound wave signals captured by electroacoustic microphones, validate the precision of the numerical calculations.

Temperature Compensation for MEMS Accelerometer Based on a Fusion Algorithm.

This study proposes a fusion algorithm based on forward linear prediction (FLP) and particle swarm optimization-back propagation (PSO-BP) to compensate for the temperature drift. Firstly, the accelerometer signal is broken down into several intrinsic mode functions (IMFs) using variational modal decomposition (VMD); then, according to the FE algorithm, the IMF signal is separated into mixed components, temperature drift, and pure noise. After that, the mixed noise is denoised by FLP, and PSO-BP is employed to create a model for temperature adjustment. Finally, the processed mixed noise and the processed IMFs are rebuilt to obtain the enhanced output signal. To confirm that the suggested strategy works, temperature experiments are conducted. After the output signal is processed by the VMD-FE-FLP-PSO-BP algorithm, the acceleration random walk has been improved by 23%, the zero deviation has been enhanced by 24%, and the temperature coefficient has been enhanced by 92%, compared with the original signal.

Analysis of Injury Metrics From Experimental Cardiac Injuries From Behind Armor Blunt Trauma Using Live Swine Tests: A Pilot Study.

Warfighters are issued hard body armor designed to defeat ballistic projectiles. The resulting backface deformation can injure different thoracoabdominal organs. Developed over decades ago, the behind armor blunt impact criterion of maximum 44 mm depth in clay continues to be used independent of armor type or impact location on the thoracoabdominal region covered by the armor. Because thoracoabdominal components have different energy absorption capabilities, their mode of failures and mechanical properties are different. These considerations underscore the lack of effectiveness of using the single standard to cover all thoracoabdominal components to represent the same level of injury risk. The objective of this pilot study is to conduct cardiac impact tests with a live animal model and analyze biomechanical injury candidate metrics for behind armor blunt trauma applications. Live swine tests were conducted after obtaining approvals from the U.S. DoD. Trachea tubes. An intravenous line were introduced into the swine before administering anesthesia. Pressure transducers were inserted into lungs and aorta. An indenter simulating backface deformation profiles produced by body armor from military-relevant ballistics to human cadavers delivered impact to the heart region. The approved test protocol included 6-hour monitoring and necropsies. Indenter accelerometer signals were processed to compute the velocity and deflection, and their peak magnitudes were obtained. The deflection-time signal was normalized with respect to chest depth along the impact axis. The peak magnitude of the viscous criterion, kinetic energy, force, momentum and stiffness were obtained. Out of the 8 specimens, 2 were sham controls. The mean total body mass and soft tissue thickness at the impact site were 81.1 ± 4.1 kg and 3.8 ± 1.1 cm. The peak velocities ranged from 30 to 59 m/s, normalized deflections ranged from 15 to 21%, and energies ranged from 105 to 407 J. The range in momentum and stiffness were 7.0 to 13.9 kg-m/s and 22.3 to 79.9 N/m. The maximum forces and impulse data ranged from 2.9 to 11.7 kN and 1.9 to 5.8 N-s. The peak viscous criterion ranged from 2.0 to 5.3 m/s. One animal did not sustain any injuries, 2 had cardiac injuries, and others had lung and skeletal injuries. The present study applied blunt impact loads to the live swine cardiac region and determined potential candidate injury metrics for characterization. The sample size of 6 swine produced injuries ranging from none to pure skeletal to pure organ trauma. The viscous criterion metric associated with the response of the animal demonstrated a differing pattern than other variables with increasing velocity. These findings demonstrate that our live animal experimental design can be effectively used with testing additional samples to develop behind armor blunt injury criteria for cardiac trauma in the form of risk curves. Injury criteria obtained for cardiac trauma can be used to enhance the effectiveness of the body armor, reduce morbidity and mortality, and improve warfighter readiness in combat operations.

Anatomical Registration of Implanted Sensors Improves Accuracy of Trunk Tilt Estimates with a Networked Neuroprosthesis.

For individuals with spinal cord injuries (SCIs) above the midthoracic level, a common complication is the partial or complete loss of trunk stability in the seated position. Functional neuromuscular stimulation (FNS) can restore seated posture and other motor functions after paralysis by applying small electrical currents to the peripheral motor nerves. In particular, the Networked Neuroprosthesis (NNP) is a fully implanted, modular FNS system that is also capable of capturing information from embedded accelerometers for measuring trunk tilt for feedback control of stimulation. The NNP modules containing the accelerometers are located in the body based on surgical constraints. As such, their exact orientations are generally unknown and cannot be easily assessed. In this study, a method for estimating trunk tilt that employed the Gram-Schmidt method to reorient acceleration signals to the anatomical axes of the body was developed and deployed in individuals with SCI using the implanted NNP system. An anatomically realistic model of a human trunk and five accelerometer sensors was developed to verify the accuracy of the reorientation algorithm. Correlation coefficients and root mean square errors (RMSEs) were calculated to compare target trunk tilt estimates and tilt estimates derived from simulated accelerometer signals under a variety of conditions. Simulated trunk tilt estimates with correlation coefficients above 0.92 and RMSEs below 5° were achieved. The algorithm was then applied to accelerometer signals from implanted sensors installed in three NNP recipients. Error analysis was performed by comparing the correlation coefficients and RMSEs derived from trunk tilt estimates calculated from implanted sensor signals to those calculated via motion capture data, which served as the gold standard. NNP-derived trunk tilt estimates exhibited correlation coefficients between 0.80 and 0.95 and RMSEs below 13° for both pitch and roll in most cases. These findings suggest that the algorithm is effective at estimating trunk tilt with the implanted sensors of the NNP system, which implies that the method may be appropriate for extracting feedback signals for control systems for seated stability with NNP technology for individuals who have reduced control of their trunk due to paralysis.

AI-Enhanced Prediction of Peak Rate of Torque Development from Accelerometer Signals

AI-Enhanced Prediction of Peak Rate of Torque Development from Accelerometer Signals

Detection of focal to bilateral tonic-clonic seizures using a connected shirt.

This study was undertaken to develop and evaluate a machine learning-based algorithm for the detection of focal to bilateral tonic-clonic seizures (FBTCS) using a novel multimodal connected shirt. We prospectively recruited patients with epilepsy admitted to our epilepsy monitoring unit and asked them to wear the connected shirt while under simultaneous video-electroencephalographic monitoring. Electrocardiographic (ECG) and accelerometric (ACC) signals recorded with the connected shirt were used for the development of the seizure detection algorithm. First, we used a sliding window to extract linear and nonlinear features from both ECG and ACC signals. Then, we trained an extreme gradient boosting algorithm (XGBoost) to detect FBTCS according to seizure onset and offset annotated by three board-certified epileptologists. Finally, we applied a postprocessing step to regularize the classification output. A patientwise nested cross-validation was implemented to evaluate the performances in terms of sensitivity, false alarm rate (FAR), time in false warning (TiW), detection latency, and receiver operating characteristic area under the curve (ROC-AUC). We recorded 66 FBTCS from 42 patients who wore the connected shirt for a total of 8067 continuous hours. The XGBoost algorithm reached a sensitivity of 84.8% (56/66 seizures), with a median FAR of .55/24 h and a median TiW of 10 s/alarm. ROC-AUC was .90 (95% confidence interval = .88-.91). Median detection latency from the time of progression to the bilateral tonic-clonic phase was 25.5 s. The novel connected shirt allowed accurate detection of FBTCS with a low false alarm rate in a hospital setting. Prospective studies in a residential setting with a real-time and online seizure detection algorithm are required to validate the performance and usability of this device.

Machine Learning Methods and Visual Observations to Categorize Behavior of Grazing Cattle Using Accelerometer Signals.

Accelerometers worn by animals produce distinct behavioral signatures, which can be classified accurately using machine learning methods such as random forest decision trees. The objective of this study was to identify accelerometer signal separation among parsimonious behaviors. We achieved this objective by (1) describing functional differences in accelerometer signals among discrete behaviors, (2) identifying the optimal window size for signal pre-processing, and (3) demonstrating the number of observations required to achieve the desired level of model accuracy,. Crossbred steers (Bos taurus indicus; n = 10) were fitted with GPS collars containing a video camera and tri-axial accelerometers (read-rate = 40 Hz). Distinct behaviors from accelerometer signals, particularly for grazing, were apparent because of the head-down posture. Increasing the smoothing window size to 10 s improved classification accuracy (p < 0.05), but reducing the number of observations below 50% resulted in a decrease in accuracy for all behaviors (p < 0.05). In-pasture observation increased accuracy and precision (0.05 and 0.08 percent, respectively) compared with animal-borne collar video observations.

User gait biometrics in smart ambient applications through wearable accelerometer signals: an analysis of the influence of training setup on recognition accuracy

Gait recognition can exploit the signals from wearables, e.g., the accelerometers embedded in smart devices. At present, this kind of recognition mostly underlies subject verification: the incoming probe is compared only with the templates in the system gallery that belong to the claimed identity. For instance, several proposals tackle the continuous recognition of the device owner to detect possible theft or loss. In this case, assuming a short time between the gallery template acquisition and the probe is reasonable. This work rather investigates the viability of a wider range of applications including identification (comparison with a whole system gallery) in the medium-long term. The first contribution is a procedure for extraction and two-phase selection of the most relevant aggregate features from a gait signal. A model is trained for each identity using Logistic Regression. The second contribution is the experiments investigating the effect of the variability of the gait pattern in time. In particular, the recognition performance is influenced by the benchmark partition into training and testing sets when more acquisition sessions are available, like in the exploited ZJU-gaitacc dataset. When close-in-time acquisition data is only available, the results seem to suggest re-identification (short time among captures) as the most promising application for this kind of recognition. The exclusive use of different dataset sessions for training and testing can rather better highlight the dramatic effect of trait variability on the measured performance. This suggests acquiring enrollment data in more sessions when the intended use is in medium-long term applications of smart ambient intelligence.

Leto: Crowdsourced Radio Map Construction With Learned Topology and a Few Landmarks

Existing crowdsourced indoor positioning systems (CIPSs) usually require prior knowledge about the site and a tedious calibration process. Moreover, they may require a large number of landmarks while ignoring the topology information that may be contained in the crowdsourced data. In this paper, we present Leto, a system that uses <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">le</b> arned <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">to</b> pology information from combined user traces to construct a radio map. Leto relies on crowdsourced WiFi and accelerometer signals only without requiring any prior knowledge about the site. Our key idea is that learned topology information can reduce the required number of landmarks, while available landmarks can transform the topology into a map. We propose a novel framework that efficiently learns the map topology by a hybrid multidimensional scaling (HMDS) algorithm and accurately rectifies the map using only a few anchors by an adaptive force-directed (AFD) algorithm. We also provide a theoretical convergence analysis of the HMDS algorithm. Experimental results on real-world datasets show that Leto can capture useful topology information and achieve significant improvements in radio map construction compared to existing systems.

Comparison of Machine Learning Algorithms for Heartbeat Detection Based on Accelerometric Signals Produced by a Smart Bed.

This work aims to compare the performance of Machine Learning (ML) and Deep Learning (DL) algorithms in detecting users' heartbeats on a smart bed. Targeting non-intrusive, continuous heart monitoring during sleep time, the smart bed is equipped with a 3D solid-state accelerometer. Acceleration signals are processed through an STM 32-bit microcontroller board and transmitted to a PC for recording. A photoplethysmographic sensor is simultaneously checked for ground truth reference. A dataset has been built, by acquiring measures in a real-world set-up: 10 participants were involved, resulting in 120 min of acceleration traces which were utilized to train and evaluate various Artificial Intelligence (AI) algorithms. The experimental analysis utilizes K-fold cross-validation to ensure robust model testing across different subsets of the dataset. Various ML and DL algorithms are compared, each being trained and tested using the collected data. The Random Forest algorithm exhibited the highest accuracy among all compared models. While it requires longer training time compared to some ML models such as Naïve Bayes, Linear Discrimination Analysis, and K-Nearest Neighbour Classification, it keeps substantially faster than Support Vector Machine and Deep Learning models. The Random Forest model demonstrated robust performance metrics, including recall, precision, F1-scores, macro average, weighted average, and overall accuracy well above 90%. The study highlights the better performance of the Random Forest algorithm for the specific use case, achieving superior accuracy and performance metrics in detecting user heartbeats in comparison to other ML and DL models tested. The drawback of longer training times is not too relevant in the long-term monitoring target scenario, so the Random Forest model stands out as a viable solution for real-time ballistocardiographic heartbeat detection, showcasing potential for healthcare and wellness monitoring applications.