Towards lightweight stress monitoring on biometric data for IoMT environments.

A review of on-device machine learning for IoT: An energy perspective

Machine Learning Algorithm Validation: From Essentials to Advanced Applications and Implications for Regulatory Certification and Deployment.

Artificial Intelligence (AI) and Machine Learning (ML) have experienced rapid growth in both industry and academia. However, the current ML and AI models demand significant computing and processing power to achieve desired accuracy and results, often restricting their use to high-capability devices. With advancements in embedded system technology and the substantial development in the Internet of Things (IoT) industry, there is a growing desire to integrate ML techniques into resource-constrained embedded systems for ubiquitous intelligence. This aspiration has led to the emergence of TinyML, a specialized approach that enables the deployment of ML models on resource-constrained, power-efficient, and low-cost devices. Despite its potential, the implementation of ML on such devices presents challenges, including optimization, processing capacity, reliability, and maintenance. This article delves into the TinyML model, exploring its background, the tools that support it, and its applications in advanced technologies. By understanding these aspects, we can better appreciate how TinyML is transforming the landscape of AI and ML in embedded and IoT systems.

With its rising prevalence and serious complications, type 2 diabetes mellitus (T2DM) is a major worldwide health burden that calls for early detection using non-invasive screening techniques. Existing screening techniques, including OGTT, HbA1c, and fasting plasma glucose, have drawbacks in terms of accessibility, expense, and invasiveness. Recent developments in heart rate variability (HRV) analysis and machine learning (ML) offer a possible non-invasive substitute for diabetes screening. Previous research on HRV-based ML models in the classification of diabetes has issues with generalizability. The objective of this study is to develop and validate ML models using HRV features: time-domain, frequency-domain, and nonlinear HRV, to improve the prediction of T2DM. The study also evaluates the developed ML model's effectiveness against existing ML models. A retrospective dataset comprising 519 individuals (261 T2DM patients and 258 non-diabetic controls) was collected from the Autonomic Function Testing (AFT) laboratory repositories. To ensure comparability of age, gender, height, and weight among groups, post-hoc matching was used. HRV features were extracted from five-minute ECG recordings using the PowerLab data acquisition system and LabChart HRV module (ADInstruments, Sydney, Australia), following the European Society of Cardiology Task Force guidelines. An 80:20 train-test split was used to train and assess MLmodels, such as Logistic Regression, K-Nearest Neighbors (KNNs), Random Forest, Gradient Boosting, XGBoost, LightGBM, CatBoost, and AdaBoost. Accuracy, precision, recall, F1-score, area under the curve (AUC) for the receiver operating characteristic (ROC), sensitivity, and specificity were among the performance indicators. GridSearchCV was used for hyperparameter adjustment to maximize model performance. The baseline characteristics of the non-diabetic and T2DM groups were similar (p>0.05). HRV analysis showed substantial decreases in the diabetic group's time-domain (SDNN - SDof Normal-to-Normal Intervals/RMSSD - RMS of Successive Differences), frequency-domain (Low/HighFrequency - LF/HF), and nonlinear (SD2 - SD of Poincaré Plot/CVRR - Coefficient of Variation of R-R Intervals) parameters (p<0.001). With a 91.2% accuracy rate and an AUC of 0.91, CatBoost outperformed other ML models in terms of prediction. LightGBM and Random Forest, which demonstrated high sensitivity and specificity, trailed closely behind. KNN achieved the highest accuracy (98.2%) and AUC (0.99), followed by Random Forest (96.4%) and CatBoost (94.5%), while hyperparameter modification further enhanced performance. CatBoost demonstrated the highest predictive performance, with an accuracy of 91.2% and an AUC of 0.91. According to correlation analysis, the most important HRV characteristics for diabetes prediction were SD2, SDRR (SD of R-R Intervals), and CVRR. This study validates the utility of HRV-based ML models for non-invasive T2DM prediction, with ensemble models like CatBoost and LightGBM demonstrating superior performance when compared to the results of prior ML models. The optimized ML model, integrated with wearable medical technology for real-time monitoring, offers a scalable, affordable, and non-invasive alternative for diabetes screening. To improve generalizability and clinical use, future studies should investigate wearable-based HRV monitoring, multimodal AI models, and longitudinal validation.

Cardiovascular diseases (CVDs) remain the leading cause of global mortality, with ischemic heart disease (IHD) being the most prevalent and deadly subtype. The growing burden of IHD underscores the urgent need for effective early detection methods that are scalable and non-invasive. Heart Rate Variability (HRV), a non-invasive physiological marker influenced by the autonomic nervous system (ANS), has shown clinical relevance in predicting adverse cardiac events. This study presents a photoplethysmography (PPG)-based Zhurek IoT device, a custom-developed Internet of Things (IoT) device for non-invasive HRV monitoring. The platform’s effectiveness was evaluated using HRV metrics from electrocardiography (ECG) and PPG signals, with machine learning (ML) models applied to the task of early IHD risk detection. ML classifiers were trained on HRV features, and the Random Forest (RF) model achieved the highest classification accuracy of 90.82%, precision of 92.11%, and recall of 91.00% when tested on real data. The model demonstrated excellent discriminative ability with an area under the ROC curve (AUC) of 0.98, reaching a sensitivity of 88% and specificity of 100% at its optimal threshold. The preliminary results suggest that data collected with the “Zhurek” IoT devices are promising for the further development of ML models for IHD risk detection. This study aimed to address the limitations of previous work, such as small datasets and a lack of validation, by utilizing real and synthetically augmented data (conditional tabular GAN (CTGAN)), as well as multi-sensor input (ECG and PPG). The findings of this pilot study can serve as a starting point for developing scalable, remote, and cost-effective screening systems. The further integration of wearable devices and intelligent algorithms is a promising direction for improving routine monitoring and advancing preventative cardiology.

The convergence of the Internet of Things (IoT) with e-health records is creating a new era of advancements in the diagnosis and treatment of disease, which is reshaping the modern landscape of healthcare. In this paper, we propose a neural networks-based smart e-health application for the prediction of Tuberculosis (TB) using serverless computing. The performance of various Convolution Neural Network (CNN) architectures using transfer learning is evaluated to prove that this technique holds promise for enhancing the capabilities of IoT and e-health systems in the future for predicting the manifestation of TB in the lungs. The work involves training, validating, and comparing Densenet-201, VGG-19, and Mobilenet-V3-Small architectures based on performance metrics such as test binary accuracy, test loss, intersection over union, precision, recall, and F1 score. The findings hint at the potential of integrating these advanced Machine Learning (ML) models within IoT and e-health frameworks, thereby paving the way for more comprehensive and data-driven approaches to enable smart healthcare. The best-performing model, VGG-19, is selected for different deployment strategies using server and serless-based environments. We used JMeter to measure the performance of the deployed model, including the average response rate, throughput, and error rate. This study provides valuable insights into the selection and deployment of ML models in healthcare, highlighting the advantages and challenges of different deployment options. Furthermore, it also allows future studies to integrate such models into IoT and e-health systems, which could enhance healthcare outcomes through more informed and timely treatments.

Internet of Things (IoT) has catapulted human ability to control our environments through ubiquitous sensing, communication, computation, and actuation. Over the past few years, IoT has joined forces with Machine Learning (ML) to embed deep intelligence at the far edge. TinyML (Tiny Machine Learning) has enabled the deployment of ML models for embedded vision on extremely lean edge hardware, bringing the power of IoT and ML together. However, TinyML powered embedded vision applications are still in a nascent stage, and they are just starting to scale to widespread real-world IoT deployment. To harness the true potential of IoT and ML, it is necessary to provide product developers with robust, easy-to-use software engineering (SE) frameworks and best practices that are customized for the unique challenges faced in TinyML engineering. Through this systematic literature review, we aggregated the key challenges reported by TinyML developers and identified state-of-art SE approaches in large-scale Computer Vision, Machine Learning, and Embedded Systems that can help address key challenges in TinyML based IoT embedded vision. In summary, our study draws synergies between SE expertise that embedded systems developers and ML developers have independently developed to help address the unique challenges in the engineering of TinyML based IoT embedded vision.

Machine learning (ML) has been embedded in many Internet of Things (IoT) applications (e.g., smart home and autonomous driving). Yet it is often infeasible to deploy ML models on IoT devices due to resource limitation. Thus, deploying trained ML models in the cloud and providing inference services to IoT devices becomes a plausible solution. To provide low-latency ML serving to massive IoT devices, a natural and promising approach is to use parallelism in computation. However, existing ML systems (e.g., Tensorflow) and cloud ML-serving platforms (e.g., SageMaker) are service-level-objective (SLO) agnostic and rely on users to manually configure the parallelism at both request and operation levels. To address this challenge, we propose a region-based reinforcement learning (RRL)-based scheduling framework for ML serving in IoT applications that can efficiently identify optimal configurations under dynamic workloads. A key observation is that the system performance under similar configurations in a region can be accurately estimated by using the system performance under one of these configurations due to their correlation. We theoretically show that the RRL approach can achieve fast convergence speed at the cost of performance loss. To improve the performance, we propose an adaptive RRL algorithm based on Bayesian optimization to balance the convergence speed and the optimality. The proposed framework is prototyped and evaluated on the Tensorflow Serving system. Extensive experimental results show that the proposed approach can outperform state-of-the-art approaches by finding near-optimal solutions over eight times faster while reducing inference latency up to 88.9% and reducing SLO violation up to 91.6%.

The integration of Internet of Things (IoT) devices in industrial applications has become viable due to advancements in ubiquitous computing that enable complex machine learning (ML) tasks on resource-constrained devices. Unlike prior approaches that rely on built-in sensors, our system utilizes externally gathered Inertial Measurement Units (IMU) data for anomaly detection. In this paper, we show that simple 1D-CNN and LSTM models on an ultra-low-power device (Nicla Sense ME) optimized for edge-based industrial anomaly detection can achieve approximately 98 movement-based anomalies (e.g., collisions and joint velocity deviations) in industrial robotic arms. We analyzed an advanced manufacturing scenario where the robotic arm performs three consecutive, distinct tasks (pick-and-place, painting, and screwdriving) and demonstrated that the proposed anomaly detection system is task-independent. We implemented these models ondevice by designing a minimal model architecture and modifying source code to minimize RAM usage and Bluetooth Low Energy (BLE) overhead. Additionally, we examined the challenges of deploying ML models in resource-constrained environments by analyzing various quantization methods and the impact of hyperparameter choices on inference time, accuracy, and memory consumption. Our approach focuses on detecting anomalies directly at the data source which enables true real-time detection with a complete edge computing framework that achieves a 10Hz data frequency and a 250ms inference time when BLE is active. Furthermore, we generated a comprehensive dataset capturing quaternion and IMU data from an industrial robotic arm over 26 hours, including various anomaly scenarios, and made the source code available on GitHub for replicability.

Stress is a mental condition that affects every aspect of life leading to sleep deprivation and various other diseases. Thus, it's necessary to analyse one's vitals to stay updated about one's mental health. This paper presents an effective method for detecting cognitive stress levels using data from a physical activity tracker device. The main goal of this system is to use sensor technology to detect stress using a machine learning approach. Individually, the impact of each stressor is assessed using ML models, followed by the construction of a NN model and assessment using ordinal logistic regression models such as logit, probit and complementary log-log. The paper uses heartbeat rate as one of the features to recognise stress and the Internet of Things (IoT) and Machine Learning (ML) to alert the situation when the person is in real danger. The patient's condition is predicted using machine learning and the patient's acute stress condition is relayed using IoT. Based on the heartbeat, a prediction of whether a person is under stress or not can be made. The paper presents a model that can predict stress levels based entirely on electrocardiogram (ECG) data, which can be measured with consumer-grade heart monitors The ECG's spectral power components, as well as time and frequency domain features of heart rate variability, are included in the model. The stress detector system takes the real-time data from the IoT device (sensor), then applies a machine learning model on the data to detect stress levels in an individual and eventually informs/alerts the individual about their stress condition. By collecting data, this system is tested and evaluated in a real-time environment with different machine learning models. Finally, a comprehensive comparative analysis has been depicted amongst the applied models with Random Forest Classifier showing the highest accuracy. The novelty of this work lies in the fact that a stress detection framework should be as unobtrusive to the user as possible.KeywordsElectrocardiography (ECG)Electromyography (EMG)Galvanic skin response (GSR)Heart rate (HR)Internet of Things (IoT)k-nearest neighbours (KNN)PhysionetRandom forestSupport vector machine (SVM)

This qualitative phenomenological study investigated machine learning (ML) model deployment challenges during the ML lifecycle using the theoretical framework of the technology acceptance model (TAM). Researchers have designed several frameworks for understanding the ML lifecycle, but those frameworks remain untested, and many ML model deployments still fail. The study’s central research question asked, what challenges do organizations face when deploying ML models in production environments? The phenomenological research design identified users’ perceptions and lived experiences deploying ML models in production environments. Data were collected via semi structured interviews with 15 ML experts. The phenomenon from the interviews was described in textural, structural, and textural-structural descriptions of participants’ lived experiences.

COVID 19-related burnout among healthcare workers in India and ECG based predictive machine learning model: Insights from the BRUCEE- Li study

This study investigates the impact of air pollution on heart rate variability (HRV), a key physiological marker reflecting the state of the autonomic nervous and cardiovascular systems. Despite growing interest, the complex relationship between environmental exposure and HRV, especially in the context of early cardiovascular disease (CVD) detection, remains insufficiently explored. An integrated real-time monitoring system was developed using Internet of Things (IoT) devices and machine learning (ML) methods to collect and analyze data from 10 healthy participants (aged 18–22) in three different environments: a controlled laboratory, an urban roadside (Al-Farabi Avenue), and a natural setting (botanical garden). Physiological signals (RMSSD, SDNN, LF, HF) were obtained using Polar H10 ECG sensors and Zhurek PPG devices, while environmental data (PM2.5, PM10, CO₂) were recorded via Tynys and Qingping sensors. Three supervised ML models—deep neural networks (DNN), random forest (RF), and XGBoost—were used to classify HRV levels based on environmental parameters. Among them, XGBoost achieved the best performance with 91.92% accuracy, 91.82% precision, and a 90.42% F1-score. The results revealed a consistent negative correlation between higher levels of PM2.5 and PM10 and reduced HRV metrics, particularly SDNN and RMSSD, indicating potential autonomic dysfunction and increased cardiovascular risk. Although CO₂ levels showed weaker associations, their influence was still noted. These findings emphasize the importance of considering environmental factors in health monitoring and demonstrate the potential of IoT and ML technologies in enabling early detection of cardiovascular stress and supporting personalized healthcare strategies.

BackgroundDermatological conditions are a relevant health problem. Each person has an average of 1.6 skin diseases per year, and consultations for skin pathology represent 20% of the total annual visits to primary care and around 35% are referred to a dermatology specialist. Machine learning (ML) models can be a good tool to help primary care professionals, as it can analyze and optimize complex sets of data. In addition, ML models are increasingly being applied to dermatology as a diagnostic decision support tool using image analysis, especially for skin cancer detection and classification.ObjectiveThis study aims to perform a prospective validation of an image analysis ML model as a diagnostic decision support tool for the diagnosis of dermatological conditions.MethodsIn this prospective study, 100 consecutive patients who visit a participant general practitioner (GP) with a skin problem in central Catalonia were recruited. Data collection was planned to last 7 months. Anonymized pictures of skin diseases were taken and introduced to the ML model interface (capable of screening for 44 different skin diseases), which returned the top 5 diagnoses by probability. The same image was also sent as a teledermatology consultation following the current stablished workflow. The GP, ML model, and dermatologist’s assessments will be compared to calculate the precision, sensitivity, specificity, and accuracy of the ML model. The results will be represented globally and individually for each skin disease class using a confusion matrix and one-versus-all methodology. The time taken to make the diagnosis will also be taken into consideration.ResultsPatient recruitment began in June 2021 and lasted for 5 months. Currently, all patients have been recruited and the images have been shown to the GPs and dermatologists. The analysis of the results has already started.ConclusionsThis study will provide information about ML models’ effectiveness and limitations. External testing is essential for regulating these diagnostic systems to deploy ML models in a primary care practice setting.

Internet of Things (IoT) is the major technology of the 4 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> industrial revolution in which various types of devices are connected together to work smartly without the intervention of humans. IoT seems to impart a great impact on our social, economic, and commercial lives. IoT applications are converting from smart home and smart me to the smart cities or smart planet. However, the large number of devices interconnected with each other by multi protocols puts the security of IoT networks on the verge of threats. Making the IoT devices more secure is also not feasible because of their limited computational power. Hence, there is a need for advancement in methods to secure IoT networks. Machine Learning (ML) models have been hot topics in security research in past years. As the IoT devices generate tons of data on a daily basis which can be used to train ML algorithms, it could be a reasonable solution to provide security to IoT systems. In this work, the main goal is to provide a broader survey of research works in the IoT security field regarding ML implementation. We briefly described the security issues in IoT networks and their impact on the privacy of important data. We then shed light on different ML algorithms and models and discussed their advantages, disadvantages, and applications in IoT individually. Moreover, the ML models currently working in IoT networks for security purposes are discussed. We also talked about the limitations of using ML models to secure the IoT networks which could provide new future research directions.