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

Minh Long Hoang,Guido Matrella,Paolo Ciampolini
DOI: https://doi.org/10.3390/s24061900
IF: 3.9
2024-03-16
Sensors
Abstract: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.
engineering, electrical & electronic,chemistry, analytical,instruments & instrumentation
What problem does this paper attempt to address?
The problem that this paper attempts to solve is: achieving non - invasive and continuous heart rate monitoring through the smart bed system. Specifically, the research aims to compare the performance of different machine learning (ML) and deep learning (DL) algorithms in detecting users' heartbeats based on the acceleration signals generated by the smart bed. ### Problem Background Currently, many wearable devices such as smart watches and chest - strap sensors have been developed for heart rate monitoring, but these devices have some limitations: - Smart watches need to be charged frequently, may be uncomfortable to wear, and users may forget to wear them. - Chest - strap sensors and other traditional methods (such as electrocardiogram ECG) are more invasive and inconvenient to use. Therefore, researchers have proposed a smart - bed - based solution to achieve non - contact and long - term heart rate monitoring. The smart bed is equipped with a 3 - axis MEMS accelerometer, which can collect users' acceleration signals and process these signals through machine learning and deep learning algorithms to detect heartbeats. ### Research Objectives 1. **Non - invasive Monitoring**: Achieve non - contact heart rate monitoring of users through the smart bed system. 2. **Continuous Monitoring**: Ensure continuous heart rate monitoring during users' sleep. 3. **Performance Comparison**: Evaluate the performance of different machine learning and deep learning algorithms in heartbeat detection and select the optimal model. ### Methods The data of 10 participants were used in the research, and a total of 120 minutes of acceleration signals were recorded. The data were divided into training sets and test sets, and K - fold cross - validation was adopted to ensure the robustness of the model. The following algorithms were compared in the experiment: - Logistic Regression - Linear Discriminant Analysis (LDA) - K - Nearest Neighbors (KNN) - Classification and Regression Trees (CART) - Naive Bayes (NB) - Support Vector Machines (SVMs) - Random Forest (RF) - Deep Neural Network (DNN) ### Results The research found that the Random Forest algorithm showed the highest accuracy among all the compared models, and its performance indicators (including recall, precision, F1 - score, etc.) all exceeded 90%. Although the training time of the Random Forest is long, this disadvantage is not significant in the long - term monitoring scenario. Therefore, the Random Forest model is considered a feasible solution for real - time cardiovascular heartbeat detection and has potential medical and health monitoring application values. Through this research, researchers hope to provide a more comfortable and reliable non - invasive heart rate monitoring method for future health monitoring technologies.