Design and prototyping of an AI-powered wearable device for continuous vital signs monitoring with intelligent alerting

Abstract

This work addresses the growing demand for affordable, wearable, and continuous health monitoring, particularly for patients, seniors, and athletes. It presents a wearable system for monitoring five vital signs: heart rate (HR), body temperature (T), blood oxygen saturation (SpO₂), blood pressure (BP), and respiratory rate (RR). All these parameters are measured using a single MAX30102 module without the need for a cuff or any additional sensors. Using advanced MATLAB signal processing, we have developed a random forest regression method, which, after simplification into a quadratic equation, estimates BP and RR with good accuracy. All recorded data are time-stamped and geotagged on a memory card, enabling patient history to support improved diagnostics. This compact, cost-effective device tracks vital signs and features intelligent alerts-audible alarms notify users of abnormalities. In critical situations, real-time vital signs and GPS coordinates are transmitted to the next of kin for emergency response. This combination of practical design and AI-powered analysis provides an effective solution for personal and clinical applications. Evaluation results demonstrate high accuracy: SpO₂ at 98.74 ± 0.99, T at 98.56 ± 0.48, HR at 95.47 ± 4.31, RR at 95.01 ± 0.96, and clinically acceptable BP estimates (systolic 94.20 ± 8.24, diastolic 92.68 ± 7.37).

Introduction

Recent advancements in wearable devices (WDs) have substantially improved vital sign monitoring, addressing many limitations of traditional methods. The global elderly population is projected to reach approximately 1.5 billion by 2050, with the fastest growth occurring in developing countries, posing major healthcare and economic challenges¹. Age-related cellular and molecular deterioration contributes to chronic inflammation and increases susceptibility to conditions such as cardiovascular disease, Alzheimer’s disease, chronic obstructive pulmonary disease (COPD), and arthritis². By 2030, COPD is expected to become the third leading cause of death worldwide; patients often experience low blood oxygen levels and hypoxemia may worsen as disease severity increases³. Among individuals over 60, the leading causes of death and disability include cardiovascular diseases (30.3%), cancer (15.1%), and respiratory disorders (9.5%)⁴. Cardiovascular diseases alone account for 43% of premature global deaths, highlighting the need for accurate and continuous health monitoring⁵.

Wearable technologies enable real-time tracking of vital parameters such as heart rate and blood pressure, supporting early diagnosis and prevention—particularly since traditional cuff-based BP measurements are unsuitable for long-term monitoring⁶. Monitoring vital signs including BP, HR, temperature, and respiratory rate is also essential for managing chronic conditions such as type 2 diabetes, where uncontrolled hypertension significantly increases the risk of stroke and kidney failure⁷. Approximately one in three to one in two adults worldwide suffers from hypertension, with men showing a slightly higher prevalence. As the leading contributor to premature mortality globally, hypertension causes millions of deaths each year; however, due to diagnostic and monitoring limitations, fewer than 30% of patients maintain adequate blood pressure control⁸.

Wearable biomedical devices play a crucial role in enhancing patient safety, particularly in Alzheimer’s care⁹ and perioperative settings¹⁰. In resource-limited environments, intelligent sensors and automated monitoring systems can improve outcomes and reduce preventable mortality^11,12. These technologies enable early anomaly detection, timely intervention, personalized treatment, and improved chronic disease management—ultimately decreasing hospital readmissions¹³. More broadly, advances in information and communication technologies (ICT), including telemedicine and wearable systems, are transforming healthcare through remote monitoring, early diagnosis, and rapid emergency response. Several major technology companies are pursuing medical-grade certifications to ensure the clinical reliability of wearable devices. Smartwatches with blood pressure monitoring capabilities, such as Omron’s HeartGuide and Samsung’s Galaxy Watch, have obtained regulatory approval within an acceptable accuracy margin of ± 5–8 mmHg, consistent with international medical device standards. Moreover, most health-oriented smartwatch models continue to comply with these standards while enhancing and expanding their monitoring capabilities¹⁴

Building on the global need for accurate and continuous vital sign monitoring, this study proposes an integrated wearable system designed to overcome existing limitations by utilizing a single sensor and advanced machine learning algorithms. Unlike conventional systems that require multiple sensors to track heart rate, blood oxygen saturation, temperature, respiratory rate, and blood pressure, the proposed design estimates all parameters via a single sensing unit. Key innovations include the indirect estimation of respiratory rate and blood pressure using AI-based algorithms developed in MATLAB, thereby eliminating the need for a cuff. The device also supports SD card data storage, efficient data processing, real-time alerts, and SMS-based geolocation for rapid emergency response. In addition , a built-in medication reminder helps users adhere to their treatment plans. This compact, integrated approach remains underexplored and may represent a meaningful advancement in wearable healthcare. To evaluate performance, measurements from 17 participants were compared with those obtained from a standard clinical device.

Background and motivation

In this section, we review key concepts, definitions, and the historical development of wearable health monitoring systems, highlighting their clinical relevance and technological progress. In the final subsection (D), the problem statement and motivation of this study are presented.

Wearable healthcare devices

Wearable systems represent a significant advancement in medical technology by allowing continuous, non-invasive monitoring of vital signs such as heart rate, blood oxygen, and blood pressure. These devices are essential for managing chronic illnesses such as cardiovascular disorders, cancer, diabetes, and neurological conditions. By enabling real-time health data collection, wearables help reduce healthcare costs and hospital stays, improving overall care quality. Typically integrated into clothing or worn directly on the body, they provide seamless health tracking¹⁵.

The market, valued at USD 80 billion in 2020, is projected to exceed USD 491.74 billion by 2032¹². Figure 1 illustrates examples of wearable devices worn on different body parts, such as smartwatches and fitness bands¹⁶. Advances in microelectronics, artificial intelligence, and wireless communication have made these devices vital for personal health management and controlling healthcare costs.

Fig. 1.

Examples of wearable health monitoring devices.

Monitoring vital signs is essential for assessing patients’ physiological conditions and promptly identifying potentially life-threatening changes. Commonly tracked parameters include heart rate (HR), blood pressure (BP), oxygen saturation (SpO₂), body temperature (T), and respiratory rate (RR), which together guide clinical decisions and reveal health trends Traditionally, nurses performed this task manually, but recent technological advances have enabled the development of intelligent wearable systems that automate and improve monitoring precision. Furthermore, as highlighted in reference¹⁷, the rise in chronic conditions such as heart failure, hypertension, and obesity—combined with healthcare staffing shortages in countries like the U.S., UK, and Canada—has increased demand for home-based telemonitoring. These systems help ease the burden on clinical facilities while enabling early diagnosis, reducing costs, and improving quality of life. Their importance becomes even more apparent during crises such as infectious disease outbreaks, when remote care is essential. For instance, these innovations gained particular relevance during the COVID-19 pandemic, promoting telehealth adoption, reducing in-person visits, lowering healthcare costs, and improving care quality¹⁸.

Despite their promise, wearable technologies still face challenges, including sensor accuracy, energy limitations, data privacy, and regulatory gaps. However, advances in Internet of Things (IoT)-enabled medical devices—including the integration of MAX30102 and infrared sensors—have significantly enhanced signal quality and system reliability, nearing hospital-grade performance. According to^1,19, as the global elderly population approaches two billion by 2050, wearable technologies will play a crucial role in addressing public health challenges. The integration of Internet of Things (IoT) technologies and Field Programmable Gate Arrays (FPGAs) has revolutionized wearable health systems by enabling real-time, high-accuracy signal processing with low power consumption (~ 0.7–1.4 W)^20–23. FPGA’s reconfigurable and parallel architecture allows efficient, adaptive detection of cardiovascular and metabolic conditions with accuracies up to 96%^20–22. Coupled with IoT connectivity, these systems support continuous, cloud- and edge-based health monitoring and personalized care^21,23–25. Advanced low-power PPG sensors further improve noninvasive blood parameter measurements while optimizing energy use^24,25. Overall, the convergence of IoT and FPGA technologies enhances computational efficiency, adaptability, and energy optimization in next-generation wearable healthcare devices^20–25.

Classification of vital signs monitoring systems

Vital signs monitoring systems are typically classified into hospital-based and wearable devices (as shown in Fig. 2). Hospital-based systems, often integrated with ventilators, represent the clinical standard for continuous monitoring, using multiple wired sensors to provide real-time data and emergency alerts. However, they have limitations such as restricted patient mobility, complex setup, high costs, infection risk, and reliance on hospital infrastructure and trained staff. These systems are also unsuitable for home or remote monitoring.

Fig. 2.

Classification of vital signs monitoring systems.

In contrast, wearable devices provide compact, wireless solutions for tracking key metrics such as heart rate, blood pressure, and oxygen saturation. They support real-time data transmission to mobile applications or cloud platforms, supporting use both inside and outside medical settings. By reducing hospital visits, they reduce healthcare costs and support remote care. Enhanced by artificial intelligence (AI) and machine learning, wearables can detect anomalies and predict health risks, playing a key role in personalized, digital healthcare and proactive health management.

Literature review

Review of hospital-based vital signs monitoring devices

Traditionally, vital signs were monitored manually through basic methods such as heart rate counting and forehead palpation for temperature—techniques used for centuries by nurses. However, increasing patient volumes and the growing complexity of care have rendered manual approaches inadequate. Today’s healthcare systems require continuous monitoring of a broader range of physiological parameters, prompting a shift toward automated technologies that have replaced mainly manual methods^15,26. The origins of non-manual monitoring trace back to 1625, when an Italian physicist introduced tools to measure body temperature and heart rate. In 1733, Stephen Hales conducted the first recorded blood pressure measurement using a glass tube in a horse’s artery, laying the foundation for later non-invasive devices^27,28. This historical trajectory led to the gradual development of modern monitoring systems, as outlined in Tables 1 and 2.

Table 1.

Several examples illustrate the evolution of hospital vital signs monitoring devices.

Table 2.

Evolution of vital signs monitoring over time.

Refs.	Year	Innovation	Description
28	Pre-1625	Manual pulse counting and forehead temperature assessment	Manual monitoring methods
26	1625	Measurement of body temperature using thermometers and heart rate with a pendulum	Early non-manual measurements
29	1707	Development of the Pulse-Watch device	Heart rate measurement
27	1733	First blood pressure measurement via insertion of a glass tube into a horse’s artery	Invasive BP measurement
27	1880	Invented by Samuel Siegfried Ritter von Basch, enabling non-invasive BP monitoring	Invention of Sphygmomanometer
29	1896	Introduction of the mercury sphygmomanometer	BP measurement
30	1903/1924	First electrocardiogram (ECG) recorded by Einthoven (nobel prize)	ECG signal recording
27	1905	Identified by Nikolai Korotkoff, allowing differentiation between systolic and diastolic BP	Discovery of Korotkoff sounds
31	1952	Introduction of the Chardiotachoscope with CRT for continuous ECG and heart rate display	Surgical monitoring
32	1956	Introduction of Electrodyne PM-65 for patient monitoring	Clinical use of production monitors
32	1966	Use of non-circular displays and nixie tube numeric indicators	Advanced display technology
32	1968	Heart rate displayed as a progress bar on CRT screens	Heart rate visualization
33	1970	Introduction of non-fading displays, isolated inputs for patient safety, and microprocessor technology	Memory monitors & digital electronics
32	1980	Development of modern modular parameter modules for patient monitoring	Modular parameter monitoring
32	1983	Bedside arrhythmia analysis and introduction of the first color physiologic monitors	Arrhythmia detection
28	1990	Development of portable monitors and cabling with flat-screen monitors and color LCD screens	Portable and flat-screen monitors
28	1994	Introduction of the first affordable digital vital sign monitor by Welch Allyn	Affordable digital monitors
32	1995	Some monitors become capable of running Windows applications	Windows-compatible monitors
32	1996	Introduction of acquisition panels replacing traditional parameter modules	Continuum of care monitors
32	2000	Integration of internet connectivity for real-time data access	IoT-enabled monitors
28	2004	Vital sign transmission via mobile phones and cloud-based analytics	Real-time remote monitoring
34	2009	Under-mattress sensors (EarlySense) measure HR & RR wirelessly	Contact-free continuous monitoring
35	2015	Embedded technologies (IoT & FPGA) in wearables—enabling real-time, adaptive, and energy-efficient health monitoring	Enabling embedded technologies
36	2018	Apple Watch delivers on-wrist ECG for public use	Consumer wearable ECG
37	2020	Massimo Safety Net & Halo ION send wireless alerts in ICUs	Smart hospital alert monitors
38	2024	i-CardiAx AI patch enables early sepsis detection via vitals	AI-based chest-patch vitals

A significant breakthrough occurred in 1994 when Welch Allyn introduced the first affordable digital vital signs monitor, which used monochrome LCDs to reduce weight. By 2004, mobile phone-based transmission of vital data enabled real-time remote monitoring. Since then, advancements have prioritized portability, accuracy, and patient comfort²⁸. Future systems are expected to evolve into AI-driven platforms that integrate physiological signals, laboratory results, and medical history to provide real-time clinical insights. Enhanced by advanced workstations and new non-invasive techniques, these systems will increasingly merge with medical imaging and signal processing to deliver more precise assessments. In the long term, nanotechnology could enable biochemical monitoring for personalized, preventive care—provided that issues of affordability and access are addressed. Therefore, as indicated in¹⁹, current research focuses on maximizing comfort and mobility outside clinical settings through continuous monitoring and intelligent algorithms for the early detection of critical conditions.

Review of wearable vital signs monitoring devices

While all wearable monitoring devices aim to track patient status, they vary considerably in features and performance. Key differences include sensor types and quantities, body placement, wearing methods, communication protocols, data display formats, processing techniques, alert systems, storage capacity, intelligence levels, user interaction modes, and personalized calibration. Notably, many commercial systems rely on proprietary technologies, which restrict independent research, performance enhancement, and feature development. Examples of such devices are summarized in Table 3.

Table 3.

Technological progress of some health watches over time.

Problem statement, motivation, and objectives

Building on the historical review and the remaining challenges in vital sign monitoring, particularly for the primary target populations—elderly individuals and patients with chronic conditions—this study addresses the urgent need for compact, accurate, and real-time wearable solutions.

Previous studies have primarily focused on single-parameter monitoring or required multiple sensors and high-cost setups, which limit portability and user convenience. These constraints highlight a major gap in developing integrated, low-cost, AI-driven wearable devices capable of multi-vital sign tracking with clinical accuracy. The growing elderly population and the increasing prevalence of chronic diseases—such as cardiovascular, respiratory, diabetic, cancer, and neurodegenerative disorders—demand continuous monitoring of key vital signs to enable early diagnosis, preventive care, and timely intervention. However, traditional hospital-based monitoring is often restricted by high costs, limited bed capacity, patient immobility, and a lack of accessibility in resource-limited areas, emphasizing the need for portable, non-invasive, and intelligent systems.

Motivated by these challenges, this study proposes a multifunctional, AI-powered wearable device capable of monitoring five essential vital signs—heart rate, body temperature, oxygen saturation, blood pressure, and respiratory rate—using a single sensor module. Innovations include indirect estimation of SBP, DBP, and RR via random forest regression, optimized sensor circuitry with advanced filtering, wireless connectivity, real-time alerts (audio and color-coded), geolocation tracking, medication reminders, and remote patient supervision. The proposed system targets vulnerable populations, including seniors, patients with chronic illnesses, athletes, and first responders, aiming to deliver reliable, continuous, and personalized health monitoring, reduce healthcare costs, minimize diagnostic errors, enable timely interventions, and provide a scalable platform for future commercialization as a smart wearable device.

Methodology

As outlined in the Introduction, this work presents the design of a wrist-worn wearable system for continuous monitoring of physiological parameters in high-risk individuals—such as the elderly, cardiac patients, diabetics, and those susceptible to stroke. The device incorporates a MAX30102 sensor, microcontroller, wireless antenna, and alkaline battery. Notably, it estimates multiple vital signs—HR, SpO₂, SBP, DBP, RR, and T—using only a single MAX30102 pulse oximeter.

Unlike conventional systems that require multiple sensors (e.g., ECG, PPG, BP), this design applies embedded algorithms to extract comprehensive health data from one sensor, thereby reducing hardware complexity and production costs. It also supports real-time data analysis and instant alerts, classifying patient conditions by urgency to enable timely intervention—crucial for vulnerable populations. The systems architecture, including hardware and software integration, is detailed in the following sections.

Hardware architecture and components of the gadget

The system integrates a compact set of electronic components (Fig. 3), centered on the ESP-Wroom-32 microcontroller, which manages signal acquisition, digital conversion, sampling, and power management. As indicated in³⁹, offering simultaneous wi-fi and Bluetooth connectivity and provides more memory than Arduino-based platforms, the ESP32 ensures stable performance.

Fig. 3.

Components of the gadget.

The MAX30102 sensor directly measures HR, SpO₂ and T, while BP and RR are algorithmically derived. For emergencies, the SIM808 GSM module sends SMS alerts and GPS coordinates to predefined contacts. A 5 V buzzer provides audible alarms, and the DS3231 RTC ensures precise timing. Data is logged on a 64 GB Micro SD card with timestamps and location information. The device is powered by a 3.7 V lithium-ion 18,650 battery. All components are managed via software developed in Arduino IDE v2.1.0, chosen for its extensive library support and hardware compatibility. This integrated architecture enables real-time health monitoring and responsive alerting, thereby enhancing patient care.

The MAX30102 sensor communicates via the I2C protocol, requiring a 5 V input while internally operating at 1.8 V for logic and 3.3 V for LEDs⁴⁰. A 100 Hz sampling rate was used to ensure precise and consistent data acquisition. PPG sensor accuracy is affected by physiological, environmental, and hardware factors⁴¹. Key influences on signal quality include sensor pressure, motion artifacts, ambient temperature (optimal range: 15–40  C), skin characteristics (e.g., pigmentation, thickness), and circulatory conditions such as diabetes. Additionally, external light interference, low oxygen saturation, short measurement durations, and hardware limitations can further degrade data quality. Longer sampling intervals (10 to 20 s) are recommended for improved signal stability. As suggested in Reference⁴², to enhance the performance of the MAX30102 sensor, it is advisable to inspect its internal circuitry, as certain production batches may require minor modifications—easily performed by end users.

According to Ref.⁴³, to address these challenges, this work adopts a hybrid filtering method combining a Moving Average Filter (MAF) and a Kalman filter (KF). MAF reduces random noise by averaging data points, while KF dynamically adjusts for motion artifacts and systemic errors. The combined MAF–KF technique enhances signal integrity and improves the accuracy of HR and SpO₂ measurements under real-world conditions.

Signal acquisition and parameter calculation using the MAX30102 sensor

The MAX30102 sensor employs PPG to measure HR and SpO₂ by emitting red (660 nm) and infrared (880 nm) light into vascular-rich tissue like the fingertip, earlobe, or lip. These wavelengths are selectively absorbed by deoxygenated hemoglobin (Hb) and oxygenated hemoglobin (HbO₂), respectively. As arterial blood volume fluctuates with each heartbeat, the intensity of reflected or transmitted light changes accordingly. These periodic optical variations are detected by a photodiode and analyzed to extract physiological parameters⁴³.

Each heartbeat increases arterial blood volume and light absorption, while diastole causes a decrease in these values. By tracking these optical pulses, the sensor determines HR by measuring the interval between successive peaks (Δt) in the signal, using the relation . Advanced processing techniques, including digital filtering and Fast Fourier Transform (FFT), enhance signal accuracy. The sensor’s low power consumption and resilience to motion artifacts make it highly suitable for continuous monitoring in wearable medical applications, as shown in Fig. 4.

Fig. 4.

How is the optical detection process engineered to ensure precise measurement of parameters: 1: Red and IR LEDs, 2: Photo-detector.

Pulse oximetry is a non-invasive method for estimating SpO₂, based on the distinct absorption properties of oxygenated (HbO₂) and deoxygenated hemoglobin (Hb) at red (~ 660 nm) and infrared (~ 880 nm) wavelengths. Using alternating emissions from two LEDs and a photodiode to detect transmitted light through vascular tissue (e.g., fingertip), the device calculates an absorption ratio that varies with blood volume during heartbeats. This ratio is used to determine SpO₂, with accuracy influenced by sensor design, physiological conditions, and the quality of signal filtering. The calculation relies on the following formula:

1

In this equation, and represent the intensities of transmitted infrared and red light, respectively. The “pulse” and “non-pulse” components refer to the periods during and between heartbeats, respectively⁴⁴. This allows precise quantification of SpO₂, expressed as a percentage (0–100%), and serves as a key indicator of respiratory and circulatory health. The MAX30102 also features an internal temperature sensor that monitors the ambient or skin surface temperature at the point of contact. This measurement, accurate to ± 1  C, supports optical calibration by correcting for thermal effects on light absorption⁴⁵. Changes in temperature can impact tissue optical properties; thus, thermal data is used to adjust signal processing and improve overall measurement reliability. While the sensor does not reflect core body temperature—being affected by factors such as peripheral blood flow and ambient conditions—it is valuable for tracking relative changes, detecting abnormalities (e.g., fever or hypothermia), and enhancing the accuracy of HR and SpO₂ readings.

PPG signal acquisition

Photoplethysmography (PPG) is a non-invasive optical technique for monitoring blood volume changes in the skin’s microvasculature, based on the Beer–Lambert law. Light emitted into tissue is absorbed or reflected depending on blood flow, with variations detected by a photodetector⁴⁶. First introduced by Hertzman in 1937, PPG underpins modern optical monitoring systems. The MAX30102 sensor integrates an LED and photodiode to capture reflected light, enabling estimation of SpO₂, vascular tone, and respiratory-induced changes⁴⁷.

This work employs a reflective PPG sensor with a near-infrared emitter (700–900 nm) enclosed in a black casing to reduce ambient light interference⁴⁸. The signal consists of a DC component (static tissues) and an AC component (pulsatile blood flow). Key waveform features—such as the systolic peak, diastolic slope, and microvascular oscillations—enable extraction of HR, HRV, SBP, DBP, and RR. To enhance accuracy, advanced signal processing methods such as adaptive filtering and optimization algorithms are applied. Figure 5 presents the characteristic PPG waveform and its key points.

Fig. 5.

PPG Chart and four key points on it obtained from the symmetry of the IR chart: 1: Foot, 2: Systolic peak, 3: Notch, 4: Diastolic peak.

Indirect measurement of SBP, DBP and RR

Continuous monitoring of blood pressure (SBP and DBP) is critical for early cardiovascular risk detection. While cuff-based methods remain the clinical standard for accuracy, they are unsuitable for continuous use due to discomfort and manual operation requirements^18,49. This has spurred the development of cuffless, wearable solutions that estimate BP from pulse signals in real-time.

BP measurement has evolved from aneroid devices to oscillometric techniques since the 1930s. However, no cuffless method has yet achieved universal validation. Challenges include accuracy, stability, and integration into healthcare systems. PPG-based BP estimation shows promise but faces interpretation difficulties and often requires user-specific calibration. Pulse Transit Time (PTT) offers better correlation but involves multiple sensors, reducing practicality^50,51. Respiratory rate (RR), another key vital sign, typically ranges from 12 to 20 breaths per minute and is affected by conditions like asthma or heart failure. Since direct RR measurement is intrusive, indirect methods using ECG and PPG signal characteristics (e.g., amplitude, frequency) are commonly employed⁵². These sensor-based, non-invasive techniques enable practical, continuous monitoring of BP and RR in daily settings.

Classification and range definition of selected vital signs

In this work, we categorize vital signs into five color-coded levels, each corresponding to a defined degree of clinical concern—an approach inspired by the National Early Warning Score (NEWS) standard⁵³. However, applying uniform thresholds can be misleading, as physiological norms vary significantly with age, gender, fitness level, altitude, and individual health profiles. For example, while athletes often exhibit lower resting HR, and elderly individuals may have higher BP ranges that are still normal for their age group.

To increase diagnostic accuracy, these thresholds should be personalized by healthcare professionals who consider medical history, lifestyle, and individual physiology. Smart monitoring devices leverage sensors to collect continuous or periodic vital sign data, visualized as time-series plots. Each data point, recorded across morning, afternoon, and night intervals, can be connected via interpolation to reveal physiological trends throughout a 24-hour cycle (Fig. 6). These visualizations are compelling for identifying unusual patterns—such as nocturnal tachycardia or episodic hypoxemia—and allow for timely interventions. The model’s color-coded logic assigns health significance to each value:

• Dark green (stable): Normal range, no action needed; device records silently.

• Light green (caution): Slight deviation; no alert, but suggests minor variation.

• Yellow (at risk): Indicates possible health risk, prompting a flashing warning and suggesting reassessment.

• Orange (critical): Dangerous state requiring immediate medical attention; the device alerts both the user and healthcare providers.

• Dark red (emergency): Life-threatening condition; triggers emergency notifications or automatic calls to caregivers.

Fig. 6.

Concept of color ranges for each vital sign (hypothetical chart).

This integrated system, combining real-time monitoring, intelligent alerts, and adaptive thresholds, empowers users to manage their health more proactively. When combined with advanced analytics, it not only improves personal health tracking but also enhances early disease detection and emergency response.

Data recording, storage, transmission, audio alerts, emergency messaging, medication reminder

Vital signs are central to innovative healthcare systems. The proposed wrist-worn device enables remote monitoring, allowing clinicians to assess patients without requiring in-person visits. It includes 64 GB of expandable memory, capable of storing time-stamped and geotagged health data for up to one year. Measurements can be configured at user-defined intervals (e.g., every 120 s), with data exportable to MATLAB or Python for further analysis. This cost-effective wearable continuously monitors HR, BP, T, RR, and SpO₂ using an embedded, non-invasive sensor. Designed for settings such as nursing homes, it ensures privacy and supports uninterrupted care. It also includes a medication reminder to help users adhere to their treatment schedule. The device features an intelligent alert system. When any vital sign reaches Alert Level 4 (orange), a buzzer and visual warning are activated. At Level 5 (red), the most critical state, it automatically sends an SMS to pre-set contacts. The message includes all vital signs, precise GPS coordinates, and a timestamp. As depicted in Fig. 7, this function enhances emergency responsiveness via real-time remote notifications.

Fig. 7.

SMS received from the gadget during user emergency conditions.

Software system design

Supporting this functionality, the device’s software system—illustrated in Fig. 8—operates within the Arduino environment. It simultaneously measures, filters, and processes vital signs using digital algorithms to enhance accuracy and reduce noise. The framework also enables data transmission to the storage card, displays real-time patient status with color-coded indicators, and configures both visual and audio alerts, thereby forming a robust and integrated health monitoring solution.

Fig. 8.

General coding framework of the designed system, consisting of five functional sections: graphical interface, vital signs measurement, emergency SMS alerts, warning system, and data storage.

User interface design

A vital component of wearable health systems is the graphical user interface (GUI), which must provide quick and easy access to critical health data. This project developed a touch-enabled color touchscreen (Fig. 9) displaying key vital signs——and medication reminders, alongside current time, date, and user details. Data are presented numerically and with a customizable five-level color code indicating stability and risk (green to dark red). Although the current prototype uses development boards and USB connections in a laboratory setting, it is designed for future miniaturization into a wrist-worn smartwatch with an internal rechargeable battery, eliminating cables. This proof-of-concept validates system functionality and establishes the groundwork for commercialization, aiming for a compact, accurate, multi-parameter wearable device with an intuitive GUI that supports continuous monitoring.

Fig. 9.

Overview of the user panel prototype (measured from two different individuals).

Dataset collection for indirect estimation of SBP, DBP, and RR

This section describes a novel approach to indirectly estimate SBP, DBP, and RR using continuous infrared (IR) data captured by the MAX30102 sensor. The raw IR signals, influenced by light absorption in biological tissues, are filtered and preprocessed to generate a PPG waveform, which reflects blood volume changes in subcutaneous tissue and forms the basis for further analysis (Fig. 10).

Fig. 10.

PPG signal based on IR.

From each PPG cycle, four key points are extracted in real time: Foot (start of systole, lowest point), Systolic Peak (highest point during systole), Notch (slight dip between systole and diastole), and Diastolic Peak (secondary peak during diastole related to ventricular filling)⁴⁸. These features are displayed numerically on an Arduino serial monitor, with Fig. 11 presenting an example of accurate detection over ten PPG cycles.

Fig. 11.

Identification of four key points in each PPG cycle.

To build a meaningful and diverse dataset, 17 volunteers (aged 16–70, both genders) contributed approximately 10 samples each. The extracted PPG features served as inputs, while actual SBP, DBP, and RR values—measured by a standard arm-cuff sphygmomanometer—were recorded as targets. The dataset, stored in Excel, contains 170 samples organized into seven columns: four input features (Foot–X₁, Notch–X₂, Diastolic–X₃, Systolic–X₄) and three outputs (SBP–Y₁, DBP–Y₂, RR–Y₃). Before training, samples were randomly shuffled to prevent overfitting, then split into 80% training and 20% testing sets to ensure model generalizability.

Random forest regression model

The random forest regression (RF) model is a versatile machine learning method that predicts continuous outcomes by averaging the results of multiple decision trees, each built independently from random subsets of data and features. This ensemble technique improves accuracy, lowers variance, and reduces overfitting compared to traditional models. Its strength lies in capturing both linear and nonlinear relationships between inputs and outputs. In this work, the RF model effectively captured a combined first- and second-order polynomial relationship—mixing linear and nonlinear terms—between the four extracted PPG features and physiological parameters, without requiring an explicit functional form. This ability to model complex patterns makes RF ideal for regression tasks with nonlinear data. The output of this hybrid model enables the definition of a quasi-analytical mathematical relation between the four inputs and each output. The structure of this equation is a second-order polynomial, generally expressed for each output (e.g., SBP) as follows:

2

where: are the numerical values of the four key points extracted from the PPG signal, represent the outputs—SBP, DBP, or RR. Moreover, are constants derived via machine learning and model fitting on training data to map inputs to outputs optimally. This approach enables accurate estimation of critical cardiac and respiratory parameters without the need for expensive or invasive tools. Furthermore, personalizing the model for individual users—training separate models per person—can improve accuracy beyond the current results, as physiological differences are reflected in the varying coefficients of the equation. In Ref.⁴⁷, the method’s error standard deviation is reported to be under 8 mmHg. Overall, this predictive framework supports the development of portable, cost-effective personal health monitoring devices competitive with emerging smart health technologies.

Evaluation and analysis of all recorded parameters

This section evaluates the proposed model’s performance in estimating systolic blood pressure (SBP), diastolic blood pressure (DBP), and respiratory rate (RR) using four widely accepted metrics: mean absolute error (MAE), root mean square error (RMSE), coefficient of determination (R2), and mean absolute percentage error (MAPE)⁵¹.

1. Mean Absolute Error (MAE): The MAE measures the average absolute difference between predicted and actual values, with lower values indicating better accuracy, defined as:

   3

Where represents the total number of samples, is the actual value (measured by the reference sphygmomanometer) for the sample, and denotes the predicted value by the model for the same sample.

• 2.
  Root Mean Squared Error (RMSE): The square root of the (MSE), retains the original data units and reflects the average prediction error magnitude:

  4

Lower RMSE values signal closer alignment between predictions and actual values.

• 3.
  Coefficient of Determination (R²): The quantifies the model’s goodness of fit, ranging from 0 to 1, with values near 1 indicating strong predictive capability:

  5

where is the mean of actual values.

• 4.
  Mean Absolute Percentage Error (MAPE): MAPE expresses prediction error as a percentage relative to actual values, making it useful for outputs with varying units or scales:

  6

Lower MAPE values indicate better accuracy, and model performance can also be simply expressed as:

7

With values closer to 100% denoting superior predictive performance.

Hyperparameter tuning of the model

This work used the random forest algorithm to estimate SBP, DBP, and RR. As a nonparametric ensemble method built on multiple decision trees, random forest requires no learning rate adjustment and is naturally resistant to overfitting. To improve generalizability and performance, hyperparameters were optimized through an extensive random search over 15,000 independent iterations. In each iteration, data were randomly split into 80% training and 20% testing sets, with the number of trees fixed at 100 and other parameters held constant.

In each run, three independent models were trained for the aforementioned outputs, and their performance was evaluated using four metrics: mean absolute error (MAE), root mean squared error (RMSE), coefficient of determination (R2), and mean absolute percentage error (MAPE). The final score for each run i was calculated as follows:

8

The highest score for each output among the 15,000 runs was considered the primary criterion for selecting the best formula.

Data-driven modeling and performance evaluation

This section presents the experimental setup, data-driven modeling results, and performance evaluation of the proposed framework. Various machine learning techniques were compared to identify the most effective model for accurate vital-sign prediction. The obtained results, statistical analyses, and performance metrics are discussed in detail in the following subsections.

Selecting the Best Model

Several machine learning models—including linear regression (LR), artificial neural networks (ANN), support vector regression (SVR), and random forest (RF)—were evaluated for vital-sign prediction. As summarized in Table 4, the random forest model consistently outperformed the other methods, achieving the highest accuracy (≈95%) and the lowest error rates for both systolic (SBP) and diastolic (DBP) blood pressure estimation. Specifically, the RF-SBP and RF-DBP models produced the best overall performance. For SBP, the random forest model achieved an MAE of 5.16, RMSE of 8.59, and R² = 0.53, demonstrating strong predictive capability and model stability. The ensemble-based architecture of RF effectively mitigated overfitting—observed in ANN and SVR models—thereby improving regression accuracy and robustness.

Table 4.

Performance comparison of machine learning models for systolic and diastolic blood pressure estimation.

Output	MAE	RMSE	R2	Accuracy
LR-SBP	8.11	10.12	0.35	92.83
LR-DBP	6.65	8.67	− 0.16	90.75
RF-SBP	5.16	8.59	0.53	95.49
RF-DBP	3.6	6.09	0.42	94.74
ANN-SBP	11.31	17.63	− 0.95	89.54
ANN-DBP	7.81	12.32	− 1.34	89.38
SVR-DBP	10.43	12.63	− 0.002	91.01
SVR-DBP	5.76	8.1	− 0.01	92.17

It should be noted that the results presented in Table 3 represent the average performance across 100 independent runs. Further methodological details of the random forest regression approach are provided in Section III-J.

Analysis and interpretation of the random forest model performance

According to the methodology described in Section III-L, the best result for each parameter was extracted from 15,000 independent runs to ensure robust and reliable outcomes. The results in Table 5, based on MAE, RMSE, R2, and relative Accuracy, confirm that the random forest model performs effectively in estimating SBP, DBP, and RR. Low MAE and RMSE values—such as 6.888 for SBP and 4.461 for DBP—indicate minimal error, which is particularly important given the clinical relevance of BP estimation. The model also achieved high R² values (0.774 for SBP, 0.777 for DBP, and 0.898 for RR), reflecting a strong fit and its ability to model the complex relationships between PPG features and physiological outcomes. Moreover, the relative accuracy reached 95.35% for SBP, 95.10% for DBP, and 97.21% for RR, suggesting reliable and generalizable performance across all targets (as shown in Fig. 12).

Table 5.

Performance comparison using four evaluation metrics for the random forest model.

Output	MAE	RMSE	R2	Accuracy
SBP	5.541	6.888	0.774	95.35
DBP	3.616	4.461	0.777	95.10
RR	0.408	0.491	0.898	97.21

Fig. 12.

Comparison of actual and estimated values for SBP, DBP, and RR using the random forest model.

While these results highlight the model’s precision, it is important to recognize that the reported values are based on mathematical approximations of input–output relationships. Consequently, slight deviations from real-world outcomes may occur, and the results should not be interpreted as the definitive performance of the final system. A more comprehensive evaluation, incorporating all processing and estimation stages, is provided in the next section to better assess the system’s practical effectiveness.

Extracting the best formula from the random forest regression model

All random forest experiments were conducted in the MATLAB environment, where the system was trained and validated. To enable implementation on our Arduino-based microcontroller, which cannot run MATLAB or advanced machine learning algorithms, the random forest model was converted into a predictive regression formula. Specifically, random forest regression was used to derive a nonlinear quadratic formula (Eq. 2) for each vital sign. By extracting the constant coefficients of these formulas for each parameter, the system can directly estimate SBP, DBP, and RR on the microcontroller. The final performance of the system using these derived formulas is presented in the following section.

Overall gadget performance evaluation

To evaluate real-world accuracy, 33 test samples (20% of the dataset) from 17 diverse volunteers were analyzed, each providing six vital signs—T, RR, DBP, SBP, SpO₂, and HR—with actual values (Column R, measured by standard Ministry of health–approved devices: Foxy U80C for BP/HR, Yuwell YK-80B for SpO₂, HL-E31 infrared for T, and manual counting for RR) compared against gadget measurements (Column M) and absolute errors (Column E) (Table 6).

Table 6.

Real, measured values, and absolute error for test samples.

ID	HR			SpO2			SBP			DBP			RR			T
	R	M	E	R	M	E	R	M	E	R	M	E	R	M	E	R	M	E
# 1	89	80	9	99	98	1	128	116	12	76	72	4	17	15	2	37.19	36.53	0.66
# 2	77	80	− 3	97	97	0	136	118	18	83	71	12	15	15	0	37.02	37.51	− 0.49
# 3	77	81	− 4	97	99	− 2	136	117	19	83	74	9	15	15	0	36.88	36.43	0.45
# 4	83	79	4	97	97	0	110	115	− 5	62	70	− 8	16	15	1	36.92	37.30	− 0.38
# 5	83	80	3	100	99	1	110	114	− 4	62	71	− 9	16	15	1	36.81	36.60	0.21
# 6	83	77	6	98	99	− 1	110	118	− 8	62	64	− 2	16	15	1	37.45	37.92	− 0.47
# 7	83	79	4	97	97	0	110	116	− 6	62	69	− 7	16	15	1	37.42	36.90	0.52
# 8	75	74	1	99	100	− 1	125	117	8	64	70	− 6	14	14	0	37.25	36.86	0.39
# 9	87	81	6	97	97	0	130	119	11	67	72	− 5	16	15	1	37.19	37.61	− 0.42
# 10	64	61	3	98	96	2	110	119	− 9	63	66	− 3	12	12	0	36.43	35.66	0.77
# 11	64	59	5	100	99	1	110	118	− 8	63	68	− 5	12	12	0	36.75	37.12	− 0.37
# 12	80	81	− 1	98	98	0	112	114	− 2	90	75	15	16	15	1	37.56	37.18	0.38
# 13	80	80	0	100	100	0	112	113	− 1	90	74	16	16	15	1	36.36	36.41	− 0.05
# 14	80	81	− 1	99	100	− 1	112	114	− 2	90	75	15	16	15	1	36.63	37.27	− 0.64
# 15	69	70	− 1	100	99	1	120	116	4	71	71	0	13	13	0	37.11	36.56	0.55
# 16	69	69	0	98	98	0	120	118	2	71	70	1	13	13	0	36.87	37.30	− 0.43
# 17	93	81	12	96	95	1	107	116	− 9	68	72	− 4	13	15	− 2	36.84	36.65	0.19
# 18	79	81	− 2	99	99	0	105	114	− 9	72	74	− 2	14	15	− 1	36.68	36.41	0.27
# 19	79	81	− 2	98	97	1	105	114	− 9	72	74	− 2	14	15	− 1	36	36.78	− 0.78
#20	79	81	− 2	96	98	− 2	101	112	− 11	72	76	− 4	14	15	− 1	37.39	36.98	0.41
#21	83	79	4	98	97	1	123	116	7	82	73	9	15	15	0	37.67	37.04	0.63
#22	83	79	4	99	96	3	123	119	4	82	75	7	15	15	0	36.87	37.32	− 0.45
#23	70	72	− 2	100	100	0	113	117	− 4	72	69	3	13	14	− 1	37.16	37.04	0.12
#24	70	73	− 3	97	96	1	113	118	− 5	72	72	0	13	14	− 1	37.1	37.46	− 0.36
#25	71	72	− 1	97	99	− 2	122	117	5	67	67	0	13	14	− 1	37.68	37.40	0.28
#26	71	73	− 2	96	96	0	122	118	4	67	70	− 3	13	14	− 1	37.67	38.18	− 0.51
#27	70	78	− 8	98	97	1	117	114	3	72	72	0	14	15	− 1	37	36.52	0.48
#28	87	80	7	98	97	1	103	115	− 12	72	72	0	16	15	1	36.87	36.37	0.50
#29	87	81	6	99	99	0	103	115	− 12	72	73	− 1	16	15	1	36.26	36.57	− 0.31
#30	82	80	2	95	97	− 2	118	117	1	78	72	6	15	15	0	36.91	36.54	0.37
#31	82	82	0	99	100	− 1	118	116	2	78	75	3	15	15	0	36.93	37.69	− 0.76
#32	82	80	2	99	99	0	118	118	0	78	73	5	15	15	0	37.03	36.44	0.59
#33	65	57	8	97	96	1	125	119	6	85	66	19	13	11	2	37.44	37.10	0.34

Across these samples, statistical metrics—Accuracy ± STD, MAPE, R², RMSE, MAE, error variance (VAR), error STD, and mean error (ME)—demonstrate high reliability: SpO₂ and T achieved accuracies of 98.74% and 98.56%, respectively, with minimal MAPE and RMSE. Other parameters (HR, RR, SBP, DBP) exceed 92% accuracy, confirming the device’s precise, stable performance and robust generalizability under variable biological conditions. Based on Table 7, the gadget demonstrates strong performance in measuring vital signs with high accuracy and consistency. Low standard deviations and near-zero mean errors across most parameters further confirm the system’s effective calibration and reliability for practical use. High R2 values for and HR further reflect the model’s strong measurement capability and generalization.

Table 7.

Statistical evaluation of the gadget’s performance in measuring vital signs.

	ME	STD	VAR	MAE	RMSE	R2	MAPE	Accuracy	Accuracy ± STD
HR	− 1.63	4.31	18.61	3.57	4.55	0.62	4.52	95.47	95.47 ± 4.31
SpO2	− 0.21	0.99	0.99	0.73	0.94	0.80	1.26	98.74	98.74 ± 0.99
SBP	0.30	8.24	67.96	6.72	8.12	0.18	5.79	94.20	94.20 ± 8.24
DBP	− 1.90	7.37	54.39	5.60	7.50	0.21	7.3	92.68	92.68 ± 7.37
RR	− 0.12	0.96	0.92	0.72	0.95	0.51	4.98	95.01	95.01 ± 0.96
T	− 0.05	0.48	0.23	0.44	0.57	0.93	1.44	98.56	98.56 ± 0.48

However, SBP and DBP show comparatively lower R² values (0.18 and 0.21, respectively), which arise from the complexity of modeling these parameters using simplified quadratic equations. While the original random forest model (Fig. 12) achieved R² values around 0.8—highlighting its strength in capturing nonlinear relationships—the transition to closed-form polynomial expressions introduces approximation errors. Given the physiological intricacies influencing SBP and DBP, second-order models may lack the flexibility to fully represent these dynamics. Despite this limitation, RMSE, MAE, MAPE, and Accuracy values remain within acceptable bounds, indicating that measurements are still dependable. This highlights the trade-off between model interpretability and measurement accuracy when simplifying machine learning outputs. Enhancing the dataset with more personalized samples from a broader range of users—varying in age, gender, and health conditions—would strengthen model training, improve generalizability, and increase measurement accuracy in real-world scenarios.

Comparison with existing methods

In Table 8, we provide a comparative overview with previous studies on blood pressure (BP) estimation using PPG signals. While making direct comparisons are challenging due to variations in data collection procedures, feature extraction methods, and evaluation criteria, we aimed to select works that aligned with our methodological framework. Specifically, we focused on studies employing machine learning models trained on features derived from PPG data.

Table 8.

Comparison of non-invasive vital sign estimation methods and accuracy.

Refs.	Year	Description	Some metrics
45	2025	Two MAX30102 sensors on back of thigh by 17 volunteers PPG signal acquisition using optical sensors	Accuracy (HR/RR) : 95.9/91.3
54	2025	Evaluation of the accuracy of a HUAWEI WATCH D2 oscillometric watch blood pressure monitor for measuring resting blood pressure by 95 persons	MAE ± STD (SBP): 7.3 ± 0.9 MAE ± STD (DBP): 4.9 ± 0.7
3	2025	Measurement SpO2 by 100 patients with COPD	Accuracy ± STD(SpO2): 93.7 ± 3.8
55	2025	Measured by the Samsung Galaxy watch 6 device, data was collected from 896 participants	MAE ± STD (SBP): 4.64 ± 4.73 MAE ± STD (DBP): 3.66 ± 3.62
8	2024	Comparison of three famous health watches (BPro(1)—Heartisanse (2)—Omron (3)) under the same conditions with 128 volunteers	(1): MAE (SBP/DBP) : 9.25/5.57 (2): MAE (SBP/DBP) : 19.74/12.14 (3): MAE (SBP/DBP) : 12.14/8.36
56	2024	Beat-by-beat estimation from PPG Hybrid Wavelet Scattering Transform (WST)—LSTM	RMSE (SBP/DBP): 15.47/11.2 MAE (SBP/DBP) : 13.38/9.53
57	2023	Non-invasive BP via machine learning (XGBoost) regression-based estimation from PPG	RMSE (SBP/DBP): 5.67/3.95 MAE (SBP/DBP) : 3.12/2.11
58	2022	Continuous monitoring of BP using LiveOne wrist-worn device Simultaneous comparison with invasive arterial line (A-line)	MAE (SBP/DBP) : 8.2/6.4
9	2021	Direct sensing via Wi-Fi microcontroller with electronic circuits By 13 healthy volunteers	Accuracy ± STD (HR): 80.7 ± 5.6 Accuracy ± STD (SpO2): 95 ± 0.022 Accuracy ± STD (T): 35.2 ± 0.77
59	2019	Estimating BP trends and nocturnal dip from PPG (wrist PPG sensor) PPG-based HR variability + pulse morphology features (LSTM algorithm)	RMSE (SBP/DBP): 8.22 / 6.55
60	2019	Machine learning based on morphological features of PPG Only PPG used (no ECG)	MAE ± STD (SBP): 8.22 ± 10.38 MAE ± STD (DBP): 4.17 ± 4.22
–		This Work	Table 6

In our study, vital signs including heart rate (HR), oxygen saturation , and temperature (T) were measured directly using the MAX3012 sensor, with appropriate signal enhancement and preprocessing. In contrast, systolic and diastolic blood pressure (SBP, DBP) and respiratory rate (RR) were estimated indirectly using a hybrid approach based on random forest regression. The final outputs were formulated through a quadratic mathematical model. Despite the inherent challenges of indirect estimation, the dataset collected in our work demonstrated satisfactory overall accuracy.

It is important to note that comparing our findings with those of other studies is inherently challenging due to substantial variations in experimental configurations. These include differences in the number and measured vital signs, sensor placement on the body, the number of participants and recorded samples, the duration of measurements, the types and quantities of sensors employed, and the methodologies used for signal processing and parameter estimation. Furthermore, the choice of machine learning algorithms and evaluation metrics varies widely across studies, making direct performance comparisons highly complex. Nonetheless, as summarized in Table 8, our proposed wrist-worn wearable system demonstrates superior estimation accuracy compared to several recent studies. Although slight deviations from clinical benchmarks exist, the system’s performance remains remarkably close to the standards defined by the Association for the Advancement of Medical Instrumentation (AAMI), which recommends a mean error below 5 mmHg and a standard deviation under eight mmHg for reliable non-invasive blood pressure monitors⁶¹. Considering the compact design and wearable nature of the device—as opposed to hospital-grade equipment this level of accuracy reflects strong potential for practical deployment in telehealth and continuous personal monitoring applications.

Conclusion

This study presents a robust, low-power, and ergonomic wearable system for continuous monitoring of six vital signs: heart rate (HR), systolic and diastolic blood pressure (SBP and DBP), respiratory rate (RR), body temperature (T), and oxygen saturation SpO₂. Primarily targeting elderly individuals and patients with chronic conditions, the device offers a compact, affordable, and non-invasive solution for real-time health assessment and preventive intervention. A total of 170 samples from 17 participants were collected, with 20% reserved for testing to evaluate system generalizability, though it should be noted that the limited sample size and demographic diversity may affect the model’s generalizability to broader populations. Experimental results demonstrate that the random forest regression model provides high accuracy and stability in estimating RR and BP, while direct measurements of HR, T, and SpO₂ using the MAX30102 optical sensor are also highly reliable. System performance reached levels close to or exceeding clinical standards (AAMI), with mean error < 5 mmHg and standard deviation < 8 mmHg, highlighting its practical applicability. A key feature of the system is its single-sensor design, which reduces weight, energy consumption, and cost while improving portability, ergonomic comfort, and ease of use. Users, regardless of literacy or technical knowledge, can easily interpret their health status using visual color-coded alerts, which are complemented by multi-level audio alerts, including an emergency alarm and SMS notification with the patient’s geolocation. This functionality enhances remote monitoring, integration with smart health platforms, and continuous on-device data storage of all vital signs. Compared with the last decade of studies, the proposed wrist-worn system demonstrates superior accuracy and stability in blood pressure estimation based on PPG signals. Despite methodological differences among studies, the system’s performance remains very close to or better than clinical benchmarks. Additional functionalities—such as medication reminders, vital history tracking, remote monitoring, and emergency data transmission—further enhance usability, patient safety, and telehealth integration. Overall, this research represents a significant advancement in wearable health technology, integrating biomedical engineering, IoT, machine learning, and intelligent hardware design to meet real patient needs. The system provides reliable, continuous, and personalized monitoring, reduces healthcare costs, enables timely interventions, and offers a scalable platform for elderly and chronic patients, ultimately supporting improved quality of life and longevity.

Future research

To build on these findings, future research should expand both the sample size and participant diversity by including individuals with specific clinical conditions (e.g., diabetes, heart failure), integrate motion and environmental sensors (accelerometers, gyroscopes, ambient temperature/humidity) for richer health assessments, and conduct long-term real-world evaluations to assess durability and stability. Linking the device to electronic health records (EHR) can improve data sharing and clinical workflow, while adaptive learning models personalized to individual physiology could boost accuracy and reduce false alarms. Finally, using metaheuristic optimization algorithms^62,63, optimizing energy consumption, ergonomic design, and adopting low-power communication technologies (e.g., BLE) are essential steps toward commercializing the system.

Author contributions

**H.S.S.** was responsible for the system’s hardware and software implementation, data analysis, hardware and software diagnostics, as well as manuscript writing and editing.—**M. A**. had overall responsibility for the research idea development and evaluation, supervised the work, and ensured the successful coordination of all project activities.—**D.R.** was responsible for system simulation, hardware procurement, and active participation in tests.

Funding

The present article is financially supported by the Research Department of the School of Medicine, Shahid Beheshti University of Medical Sciences (Grant No. 43010314; ethical approval under code [IR.SBMU.MSP.REC.1403.213]).

Data availability

All data generated or analyzed during this study are included in this published article. Additional raw data supporting the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval

This study was reviewed and approved by the Ethics Committee of Shahid Beheshti University of Medical Sciences under the approval code [IR.SBMU.MSP.REC.1403.019]. All participants voluntarily participated in the sample collection with full awareness and informed consent. This study confirms that all experiments were performed in accordance with relevant guidelines and regulations.