Explore the intersection of big data and IoT analytics in biomedical and healthcare technologies, emphasizing the application of machine learning and AI techniques for remote diagnostics and telemedicine. They discuss the adaptability of AI-based telemedicine in advancing healthcare (Banerjee et al., 2020; Kim et al., 2020). Highlight the transformative impact of AI and IoT technologies in healthcare, introducing innovative diagnostic tools and care strategies. Underscores IoT’s role in healthcare, facilitating remote monitoring, smart sensors, and medication delivery to enhance patient care despite IT management challenges. Focus on IoT systems in smart homes for elderly care, Narejo (2020) and Sokullu et al. (2020) describing how environmental sensors provide context-aware monitoring to ensure safety and provide early warnings for individuals with memory issues. They present a low-cost prototype integrating emergency features for handling critical situations. IoT based waste management system is one example which empowers the smart home-based system in this technology-oriented era (Addas et al., 2024).

Regarding the growing need for eldercare because of the world’s aging population, Stavropoulos et al. (2020) draw attention to ambient assisted living technology such as wearables from the IoT and sensors that allow for remote monitoring and assistance. After reviewing case studies, they describe current trends and prospects in effective eldercare technology, grouping solutions by aims, durations of experiments, IoT technologies, and result measurements (Hassan et al., 2020). A Spark streaming framework for real-time health status prediction, with an emphasis on real-time tracking in healthcare, using wearable sensors and mobile apps. They emphasize potential in diabetes treatment. Their method combines streaming machine learning models to achieve high accuracy in health data processing. IoT-based data mining and AI are supported by the healthcare industry to solve issues such as false and incomplete medical records (Muthu et al., 2020). They suggest the Generalized Approximate Reasoning-based Intelligence Control (GARIC), which emphasizes tailored health alerts and therapies, for the study of patient data and illness prediction. Investigate how technology may help reduce senior loneliness amidst the COVID-19 epidemic. They recommend judicious use of technology in crisis communications and care initiatives to help older individuals feel less alone. They emphasize the value of easily available and adaptable treatments for different forms of loneliness and demographic factors (Conroy et al., 2020).

Authors discussed the increasing challenges of aging populations, emphasizing the integration of AI and IoT in assisted living and healthcare monitoring for older people. They provide a comprehensive overview, comparing techniques and application scenarios while discussing the benefits and drawbacks of these technologies (Qian et al., 2021) explore the shift towards personalized healthcare systems using AI and machine learning, focusing on disease diagnosis, health monitoring via wearables, and assistive frameworks like social robots. They review current smart healthcare systems, highlighting integration designs critical for intelligent healthcare solutions (Nasr et al., 2021).

Recent advancements in healthcare management systems (HMS) have been significantly influenced by the convergence of blockchain, IoT, and AI technologies. Individualized approaches have replaced traditional hub-based systems (Junaid et al., 2022). However, heterogeneity of devices, disparate IoT designs, and a lack of competitively priced smart sensors are barriers to interoperability and data integration in the deployment. Despite the benefits of telehealth for senior care, privacy issues, especially in this context, provide substantial obstacles that must be addressed with improved privacy safeguards such as informed consent to increase adoption rates. Innovations in AI and IoT are also essential to improving telemedicine, especially in fall detection systems that use cutting-edge algorithms and sensor technology (Pool et al., 2022; Wang et al., 2022). In addition, the shifts in Chinese society on senior care. In addition, the shifts in Chinese society towards elder care underscore the stress placed on caregivers and the necessity of all-encompassing methods to assist both the elderly and their careers (Ai et al., 2022).

Recent studies show that edge AI is boosting decision-making and operational efficiency without requiring extensive infrastructure, transforming a variety of industries. Telepresence robots have become essential tools during the COVID-19 pandemic, helping to overcome psychological barriers to human interaction and ease remote connections and healthcare delivery (Naseer et al., 2023c,d; Sivabalan and Minu, 2022). The development of smart healthcare systems is leading to the creation of “intelligent hospitals” of the future through applications such as non-contact health screenings and intelligent hospital guidance. A critical factor in the digital transformation of healthcare systems is the incorporation of AI, with particular attention paid to data infrastructure, system integration, and real-world applications. Mobile healthcare apps offer practical solutions for medical record-keeping and patient communication, though concerns persist regarding privacy and data security, prompting discussions on regulatory frameworks and risk mitigation strategies (Ding et al., 2022; Kumar et al., 2023; Monlezun, 2023).

Telepresence robotics that involve communication delays contain certain difficulties that are solved using a combined approach that uses deep reinforcement learning methods including double deep Q network (DDQN) and gated recurrent units (GRU) methods that help to control depending on time intervals (Naseer et al., 2023a,b). The appropriateness of wearable smart sensors in complementing public health projects via suitable technology incorporation has been shown revealing that these sensors are being investigated for disease control and surveillance of vital signs in epidemics. The legal and ethical issues of edge AI in healthcare point out some regulations and the possible ways to apply them responsibly in several domains. Despite challenges in data management and legal systems, the adoption of AI in healthcare leads to enhanced medical services regarding diagnosis and personalized treatment programs (Das et al., 2023; Mohammadzadeh et al., 2020).

In recent years, GANs have gained significant attention for their ability to generate realistic data across various applications, including computer vision and natural language processing. Zhang et al. (2023) proposed a Robust Generative Adversarial Network (RGAN) that enhances the generalization capabilities of GANs by promoting local robustness within the training sample neighborhood, thereby addressing common issues of instability and poor generation quality associated with traditional GANs. Cai et al. (2021) provided a comprehensive survey on GANs, highlighting their applications and the challenges faced in privacy and security contexts, which further emphasizes the versatility and importance of GANs in modern research. Additionally, Zheng et al. (2019) introduced the Vehicle Synthesis Generative Adversarial Networks (VS-GANs) framework, which effectively generates annotated vehicle images from remote sensing data, significantly improving vehicle detection performance in high-resolution images. This post-modern technology involving AI, which is increasingly used in the medical and health fields, enables the analysis of biological aspects and health conditions by interaction and correlation (Shaban-Nejad et al., 2022). AI analyzes findings using a variety of modalities. Particularly in remote patient monitoring and chronic disease management, wearable health technology has greatly revolutionized healthcare by enabling proactive health management through accurate biometric tracking and real-time monitoring. Detailed insight of AI and IoT in processing the large volume of data concerning health has enhanced the challenging aspects of data processing and ensured secure and integrated healthcare infrastructure at a large scale (Mullankandy et al., 2024; Tariq, 2024). These advancements highlight the transformative potential of AI-driven technologies in enhancing health monitoring, decision-making, and data management in modern healthcare systems.

While previous works have presented discussions on the transformative potential of GANs in healthcare applications, most of them usually forget to discuss some of the key challenges that may be evident in real-world applications. Additionally, GAN is susceptible to mode collapse—a certain model generates limited variations in generating data, which can challenge the diversity needed for analytics associated with robust healthcare. On the other hand, another critical issue is privacy concern in IoT-enabled smart homes, which remain explored to a lesser extent—one that concerns secure transmission and, at the same time, storage of sensitive health data. It comes to show that so far, the frameworks are able to meet the demands and privacy implications of IoT ecosystems while raising GAN performance. Precisely, we also focus on such limitations of integrating various robust privacy measures, specifically anonymization and encryption, taking into account a scalable architecture towards personalized elderly care within this paper.

GANs are a core component of the proposed system, enhancing its ability to address data scarcity and privacy concerns. The GAN architecture comprises a generator and a discriminator network:

Architecture:

By leveraging GANs, the system not only addresses data limitations but also enhances privacy and overall performance, making it a robust solution for IoT-enabled elderly care.

The components of embedded systems for hardware implementation involve using microcontrollers such as Arduino for data aggregation, health/environmental sensors for monitoring health/environment, and wireless modules such as Raspberry Pi for transmitting collected data to central units. Such a setup as shown in Figure 3, helps in the free flow of data or information which is vital for the kind of care delivery that is real-time. In validation, great concerns are taken in ensuring that the data is of quality, and model training with cross-validation technique heavily adheres to maximize accuracy and reliability. In summary, the system can be beneficial in Intelligent Homes to monitor the real-time health condition and status of elderly patient care and safety while promoting elder’s home liveliness.

The Perception Layer is the basic layer in this system, including sensors such as wearable devices tracking health indications of one’s body—for example, pulse and body temperature—and environmental sensors tracking the humidity, temperature, and gas levels of one’s surroundings. These sensors continuously feed in inputs to form a basis for real-time processing of data and decision-making. Starting the chain is the Perception Layer which includes wearable sensors, environmental sensors, and smart objects. Smartwatches and medical-grade wearables track body metrics like pulse, blood pressure, body temperature, and others. Several parameters of the environment include the temperature, the level of humidity, and the number of certain gases, thus, monitoring the total condition of the elderly person. Cameras and motion detectors the so-called smart devices facilitate how falls are detected and what activities the patient undertakes. Data gathered from sensors is sent to a central node with the help of Bluetooth or Wi-Fi and forms the basis of the Communication Layer. From the central hub, data is then directed to an IoMT cloud for archiving and processing in real-time. This layer makes certain that there is proper transfer of data from sensors to the cloud computing framework. The core of the system belongs to the Processing Layer where the pre-processing of data, application of the AI Model, and generation of insights occurs. Preprocessing of the data entails data cleansing, normalization, and segmentation to make it suitable for the AI algorithms. For health anomaly detection, WANN and for environmental hazards, CNNs are used from the processed data. The Application Layer also involves the management of interactions with the user and the alerts. An end-user interface through mobile apps or web interfaces for the caregivers or other medical staff makes it possible to get real-time updates and get alarms or notifications in case of the occurrence of an event of concern to the system. Information is passed on through short message service, email, or a voice call; help is summoned in case of an emergency. Thus, Algorithm 1 describes the implementation of layers, namely Perception, Communication, Processing, and Application, would ensure interoperability of the systems. The computer processing of data is a distributed process with low latency in nodes of wearable devices and central units while the heavy data processing and long-term data storing is done in the cloud. VP & ACM also provides various tiers in the architecture to enhance the system performance and the scale of elderly care delivery.

The proposed system will adapt over time through continuous feedback loops and reinforcement learning mechanisms. User-specific data, such as health metrics and behavioral patterns, dynamically fine-tune the system’s predictions and interventions. For example, if a user’s activity patterns change due to a temporary condition, the system adjusts its alerts and thresholds to minimize false positives. Reinforcement learning enhances this further with the ability to reward the model for good predictions and penalize it for bad ones, hence improving the model’s ability to make better decisions over time. Examples of personalized functionality include tailored prompts, such as reminding a user to drink more often when high levels of activity are detected, or adjusting fall detection thresholds based on observed walking stability. Such adaptive interventions ensure that the system remains responsive to individual needs and hence more effective in naturalistic settings.

The proposed system integrates a variety of IoT hardware components to ensure comprehensive monitoring and real-time data processing:

Wearable Sensors

The different privacy protection methods will be developed to amass the ethical considerations of the study. The proposed system will apply anonymization techniques that will remove PII and replace it with unique, non-traceable identifiers. Besides this, AES-256 shall be applied at transmission and storage to ensure data handling securely. It also utilizes GANs to synthesize artificial datasets representative of real health data for training models without the use of actual information from users. The RBAC limits access to such data by regulating personnel with only authorized roles, ensuring ethical standards and improving security in this IoT-enabled framework. These measures together address all issues on privacy and ensure ethical handling of health information.

For the research, GANs for data synthesis, CNNs for image analysis, and WANNs for time-series data were chosen after careful consideration against various options. While RNNs and LSTMs are widely used in time-series data for the capabilities they provide in capturing temporal dependencies, WANNs had to be chosen for this problem because of their computational efficiency and the ability to process high-frequency signals that will be critical for real-time health anomaly detection. GANs were adopted for synthesizing data because it generates high-quality, realistic data with the ability to address the class imbalance problem, which is not possible with traditional oversampling techniques. In image analysis, CNNs have been chosen because CNNs are robust in representing spatial features, which in turn provide very good efficiency in detecting falls and ambient conditions. These model selections are therefore in line with the requirements of the system in real-time performance, accuracy, and adaptability over the IoT-enabled framework.

During system development, machine learning frameworks, including TensorFlow and PyTorch, were utilized to implement and train the GANs and other AI models. For data analysis, AI-driven statistical tools such as MATLAB were used to evaluate performance metrics and generate confusion matrices. No Generative AI tools has been used in developing or writing this study.

The simulation tests are conducted to test the performance of the new AI model in aspects regarding accuracy, precision, recall, F1-score, false positive rate, and false negative rate. The results are summarized in Table 1.

For many weeks, elderly people volunteers’ houses were utilized for field testing to evaluate the system’s practical performance. The outcomes can be shown in Table 2.

The overall system performance metrics of the IoT-enabled smart home for personalized elderly care demonstrate high effectiveness and reliability, as shown in the attached chart. The system achieved an accuracy of 98.0%, indicating a high rate of correct identifications for both true positives and true negatives. With a precision of 95.0% and a recall of 94.0%, the system effectively minimizes false alarms while ensuring that critical events are detected. The F1-score of 94.5% balances precision and recall, reflecting comprehensive performance. Additionally, the response time of 2.4 s underscores the system’s capability to generate timely alerts, crucial for immediate interventions in emergencies. These metrics highlight the system’s potential to provide robust continuous monitoring and prompt response, enhancing the safety and well-being of elderly individuals through the integration of IoT and AI technologies.

The subsequent decline highlights areas needing improvement, particularly in maintaining consistent performance over time. Despite this, the system maintained high overall metrics, demonstrating its effectiveness in providing reliable and continuous monitoring for elderly care. Further efforts will focus on addressing the causes of performance degradation to ensure sustained high uptime and reliability. Table 12 provides a comprehensive comparison of the proposed system’s performance metrics with existing solutions.

Extensive field tests were conducted to evaluate the system’s performance in diverse real-world scenarios. The tests involved 50 elderly users in smart home environments over a 6-month period, capturing data across varying environmental settings and user behaviors. The following key results were observed:

Environmental Variations:

These findings highlight the robustness of the proposed system, while also identifying areas for improvement, such as advanced noise filtering techniques and improved user training for optimal sensor placement. By validating performance under real-world conditions, this study ensures the practical applicability and scalability of the system in diverse scenarios. Feedback from 50 elderly individuals and 20 caregivers over a six-month field test highlighted the system’s usability and practical benefits. Among elderly users, 85% found the system easy to use, 90% reported an increased sense of security, and 80% felt it supported their independence. Caregivers noted a 75% reduction in workload due to timely alerts and actionable insights, with 88% expressing confidence in the system’s reliability. Challenges included occasional difficulties in attaching wearable devices and the need for more customizable alert thresholds. Overall, the feedback underscores the system’s usability, effectiveness, and areas for refinement to enhance user experience.

The proposed system exhibits practical robustness by comparing the simulation and field performance. While simulation results showed slightly higher metrics due to their controlled conditions, field tests revealed that the system is able to adapt to real-world scenarios. For instance, the fall detection system resulted in high accuracy of 94.1% and a recall of 89.5% in field conditions, hence proving efficiency in picking up true events even under changing environmental conditions. Similarly, health anomaly detection and environmental monitoring were able to recognize critical health risks and hazards with reliable performances of 93.6 and 95.8%, respectively. These results underpin the practical applicability to offer timely and accurate intervention in elderly care in IoT-enabled smart homes. As it can be regarded Figure 12A, the scores of the system during the actual field are a bit lower than the outcome of the simulation, but the system still boasts relatively high accuracy and reliability in the real environment. The performance parameters are expected to drop because the performance of an algorithm or a model is bound to be different in real-life conditions, which are highly stochastic and liable to fluctuations.

The fall detection system had the best performance with an accuracy of 99% as shown in Figure 12B and it was F1-score which gave the overall rating of the model. 1%, precision of 98.9%, recall of 98. First, the precision was identified to be 91%, the recall 9%, and the F1 score 98.9%. This of course is helped further by the response time of 2. The choice of 5 s allows for early alert creation, thus meeting the goal of raising the awareness of caregivers as soon as possible.

They also showed satisfactory results in the health data anomaly detection operation where it makes an accuracy of 98 percent. 0%, precision of 97.8%, recall of 97.8% and an F1-score of 97.8%. The response time of 2. As illustrated in Figure 12C, the Δ P of the new design is less than the original design with a range of 0. This is about 8 s, which is ideal because it will in a way facilitate timely alerts in the process of caregiving.

Regarding the specific goals, the environmental hazard detection system had a 95 percent accuracy. 8%, precision of 93.8%, recall of 90. accuracy of 0%, a Precision of 86% Recall of 0%, and an F1-score of 91.8 within the minimum response time of 2.0 s. In these metrics, the system’s capacity to quickly identify adverse states and notify the caregivers is well illustrated in Figure 12D.

The comparative analysis with the existing solutions reveals a + 52% increase in accuracy and precision, a + 62% increase in recall, and improvements to the response time. The proposed system yielded higher results than the pre-existing systems in all forms of case study, it corroborated the maximum capability in giving correct and timely alerts noteworthy for elderly care as shown in Figure 12E.

The absolute performance parameters of the system can be marked as relatively high – an accuracy of 98.0%, precision of 95.0%, recall of 94. It yielded an accuracy of 0%, and an F1 score of 94.5%, with the tested response time being 2.4 s, which enhances the system’s ability to monitor the situation continuously and give a quick response. The availability and reliability measures of the system that have been discussed over a period of 3 months give an account of how sound and strong the system is and what portions require improvements, except for a trivial dip as shown in Figure 13.

Various real-world scenarios were put into test, such as the quality of sensor data or the level of environmental noise. The fall detection system, for example, performed well in high-quality sensor data scenarios with an accuracy of 99.1%, while under conditions of low-quality or noisy sensor data, the accuracy slightly dropped to 94.3%. For example, in health anomaly detection, the system showed a consistent recall of 97.8% under normal conditions but a slight fluctuation of ±2% under highly noisy environments. In environmental monitoring, the model shows a decrease in specificity from 98.8 to 95.2% under fluctuating humidity or high background interference conditions. These findings have underlined the strength of the model in terms of adaptability and robustness, hence giving good insight into its reliability in various real-world scenarios.