Sensors are a pervasive piece of technology that might potentially enhance human senses by translating complex physical ideas into measurable data. However, most modern sensors are embedded deeply within commonplace items. However, although sensors were formerly the primary program between the real world and human concept, data from several sensors is now routinely integrated and prepared, adding a layer of indifference to the process [1]. Examples of devices that benefit from sensors include smartphones, where the average user likely isn’t thinking about the raw data provided by the device’s microphone, CCD (Charge-Coupled Device) camera, MEMS (Microelectromechanical Systems) accelerometer, GSM (Global System for Mobile Communications)–modem [2], and many more. Pervasive or ubiquitous processing is just one example of how some futuristic views of computing go even further [3]. Common to most of these is the incorporation of some sort of electronic user interface into the physical environment, rendering commonplace objects interactive. Small, cheap computers that can communicate with one another and sense their surroundings are the ideal building blocks for this interface. As a result of these factors, the novel advertising concept of Wireless Sensor Network (WSN) has emerged. WSN consists of distributed stuck systems [4]. The widespread applicability of the WSN- concept suggests that there is likely to be significant interest in the related technologies. An ideal WSN would allow for the deployment of a large number of autonomous sensor nodes, each of which would tailor its settings to its surrounding nodes and the overall network structure [5]. Nodes collect data by sensing their surroundings and processing it locally before sending it, or an extract of it, to a network sink for further processing and eventual readout. While data transmissions are expected to take the most direct route to their destinations, administrators can control the entire network as though it were a single, massive gauge. Power consumption and efficiency become equally crucial while attempting to ensure the dynamic condition of the detecting device in a remote place [6].

On the Internet of Things (IoT), there is an increase of environment-sensing and -monitoring devices, leading to a crisis of insufficient energy supply. The network’s lifetime and power dissipation are both impacted by the IoT routing protocol’s energy consumption [7]. The key problem is ensuring steady data transmission from the sending node to the receiving Base Station (BS). Industrial control systems, motion sensing, and environmental monitoring are all components of WSNs. WSNs can take on multiple forms, employ various technologies, and communicate via various network topologies, making their appearance highly application-specific [8].

Environmental monitoring (EM) is typically associated with studying the natural world; however, the term can refer to any kind of environment. Observation and data collection are the foundations of EM because they allow us to learn more about the natural world around us [9]. Physical science, biological study, and chemical analysis are all components of environmental monitoring. However, the actual number of technological areas to explore increases as more systems are involved, particularly for data collecting [10]. With a growing global population as inspiration, it’s clear that environmental monitoring encompasses more than just data collection and analysis. The role that EM plays in revealing the limitations of EM and demonstrating the consequences of human behavior on the environment is crucial [11]. In addition to environmental science, popular applications include protecting drinking water supplies, treating radioactive waste, monitoring pollution, safeguarding natural resources, counting and forecasting weather, and keeping tabs on animals as shown in Fig. 1.

By analyzing a statistically significant sample of the environment, environmental monitoring attempts to pinpoint the state of the world around you. Since the sharing of information like this is crucial to EM [12]. Concerns about the invasiveness of the measuring program, the precision of the sampling, and the ability to store samples are all addressed by the data acquisition system. The degree to which these worries manifest depends, on the one hand, on the context in which they are used; for instance, on the sensitivity of the observed physical value that is to be affected externally [13]. Grab samples, (remote) sampling stations and remote sensing are examples of common sampling techniques. Grab sampling entails taking an instance out of context for the sake of further study. Although this is not done as frequently as it once was due to technical advancement, obtain sampling is still employed in select complete circumstances to allow for special samples or even more complex investigations than is possible in the field [14]. The presence of humans adds not only to the expense and inconvenience of collecting samples but also to the invasiveness of the process. Sampling stations are sensor devices that are placed across the study area and updated at regular intervals [15]. Although WSNs are commonly used for gathering environmental data in real-time via sensors, the system is not without its flaws. The issue of data validation has long plagued this field as any invalidated data will provide inaccurate forecasts and may lead to questionable decision-making, lowering the overall performance and reliability of a monitoring system [16].

The focus of the study is on WSNs and their use in environmental surveillance research. The authors note that WSNs have received a lot of attention since they can be put in difficult-to-reach areas to perform ad hoc remote sensing and communication [17]. However, there have been issues with WSN development that have prevented them from reaching their full potential. The research focuses on two major issues: the validity and trustworthiness of data, and the efficiency and effectiveness of energy consumption. Since the sensing environment can introduce flaws and inaccuracies into the sensed data, the authors stress that caution is warranted while using it [18]. Therefore, validating data is essential for making sound inferences and choices. The study also highlights the significance of keeping sensing systems online at far-flung locations while accounting for power limitations and optimizing energy consumption [19].

The study suggests a smart framework using an AI algorithm to validate data in far-flung places, which should help overcome these obstacles. The goal of the framework is to detect a wide variety of errors, such as “spike” errors, “out of range” errors, “outliers,” and “stuck-at” errors. When compared to other algorithms, the proposed one achieves a 97% performance boost in terms of validating input data [20]. WSNs are widely used for remote environmental monitoring. However, they face two critical challenges—validating the reliability of sensed data and optimizing energy consumption. The remote sensing environment can introduce errors into the collected data. Therefore, it is crucial to validate the data before using it for analysis and decision-making. However, existing data validation algorithms have limitations in detecting multiple error types [21]. Furthermore, maintaining sensing system operation with limited power in remote locales is difficult. Therefore, this research aims to develop an AI-based framework to address these issues in WSNs for environmental monitoring. The key objectives are:

Design a smart system to validate sensor data in remote areas by detecting various data faults like outliers, stuck-at errors, etc.

The proposed approach will detect a wide range of data errors through advanced AI techniques. This can significantly enhance the reliability and accuracy of the sensor data collected from remote locations. The energy-efficient IoT-WSN architecture will further optimize power utilization.

This research aims to develop an AI-based system to validate sensor data in remote areas and improve the energy efficiency of wireless sensor networks for environmental monitoring applications.

EM is frequently associated with studying the natural world; however, the term can refer to any kind of environment. Observation and data collecting are the foundations of EM, allowing for a deeper comprehension of the natural world around us. Physical science, biological study, and chemical analysis are all components of environmental monitoring [22]. As additional data-collecting systems are integrated, however, the actual number of technological areas of research increases. Given the world’s rapidly expanding population, it’s clear that EM is about much more than simply gathering information about the environment; it’s a comprehensive science with profound implications for human existence since our surroundings have a major impact on how we act and what we eat. Weather monitoring and prediction is the most important application because it allows us to protect our crops from floods and droughts and keep us from getting sick from drinking contaminated water [23]. EM is also used to keep an eye on waste material, especially radioactive waste, and to treat the environment in densely populated areas [24].

The sensor nodes in EM collect data about the surrounding environment, send it to the base node or sin node for validation, compile the data, and send it to the server, where it is displayed to the customer and potentially used for things like forecasting and prediction [25]. For environmental sensing, various sensors are linked to a processing node, typically a microcontroller, which controls and gathers data from the sensors and sends it to a corresponding sink node. However, the sensor node requires power to function, and this presents a problem in areas where power generation and storage are problematic. Grab sampling entails taking an instance out of context for the sake of further study [26]. While technological advancement has made this less common, it is nevertheless done on occasion. In these circumstances, obtain sampling is employed to allow for a more specialized sample or more complex investigation than is feasible in the field. Human involvement in the sampling process adds not only additional expense and time but also significant invasiveness. Sampling stations are sensor devices that are placed in the relevant environment and continually monitor it at regular intervals [27]. Drought in the United States in 2012 is only one example of how the value of investing in environmental monitoring has increased in recent years. This is addressed by utilizing an innovative strategy for environmental monitoring and drought forecasting. As sensing technology improves and low-cost microcontrollers become more widely available, systems like this one, which use infrared rays to gather data via remote image sensing and a geographic information system to investigate erratic air and water conditions, are becoming increasingly attractive [28]. It was suggested that a network-based application be used to keep tabs on the conditions inside a structure. The architecture of the running system is low energy. Pre-relapse System on Chip (SOC) machine wireless detectors that connect to the internet via Wi-Fi standards [29].

A small number of recently proposed solutions have centered on wireless sensing tactics and Geographic Information System (GIS)-based technologies that allow for remote monitoring of the environment [30]. Many of these approaches, however, only address the issue insofar as it pertains to data monitoring and collection. The collected data must be analyzed, and the results must be communicated to the appropriate people via timely alerts if the nagging problem is to be addressed in its entirety [31]. With the help of a reader antenna and a cluster computer, the gadget can gather RF energy, and its process has been proven in testbeds and experimentations. The look attained a gearbox mixture of up to 10 and 20 m while operating on battery power alone, respectively [32]. If the web host PC is connected to the Internet, the proposed system may be built into an IoT-based solution that can be used to keep an eye on the weather in the great outdoors. In Urban Water Systems (UWS), sensors are commonly employed to gather information and data. Standard on-line sensors, including dissolved oxygen, flow, level, oxidation-reduction potential, and suspended particles, are always installed in small Wastewater Treatment Plants (WWTPs) [33]. Larger WWTPs often use both off-line and on-line sensors [34] while smaller WWTPs typically only use off-line analysis. Most sensors in the UWS are subjected to extreme conditions, such as high temperature (for example, in thermal hydrolysis), humidity, and a corrosive environment, which can cause fouling or damage to the sensors and reduce their accuracy and reliability [35]. If the sensing signal is employed in control systems or if the sensor, like ammonium NH4+, is used for decision-making (manually or automatically), then its failure will affect the performance of the system and could cause substantial damage. The addition of sensors in a closed loop increases the possibility of error cascading through the system [36]. Therefore, safe and dependable operations of the UWS may be ensured through the early detection, diagnosis, and prognosis of failures in sensors. According to [37], degrading failure (bias, drift, and gain) and sudden failure (abrupt, noisy, and random) are the two main categories of sensor defects. The Dissolved Oxygen (DO) sensor is an integral part of the WWTPs’ aeration control system, which is critical to ensuring the plants’ safe and reliable operation [38]. Since the majority of the WWTP’s energy budget goes towards powering the aeration process, there is good reason to keep a close eye on it. Reference [39] employed principal component analysis to identify common DO sensor failures. All the aforementioned defects were simulated in a system consisting of Anaerobic-Anoxic-Oxygen (A2O) reactors to validate bias, drifting, loss of accuracy, and even complete failure, but not to validate in real data [40]. All DO sensor problems were successfully identified, and the performance of the proposed approaches was evaluated in terms of speed, efficiency, and precision [41]. This paper discusses challenges in 5G-enabled IoT, including data reliability issues. It highlights how inaccurate or invalid data can lead to incorrect decisions and system failures. The authors examine data validation challenges in IoT-based intelligent transportation systems, emphasizing how invalid data can compromise safety and efficiency [42]. This comprehensive review covers security and trust issues in IoT, including the critical need for data validation to ensure system integrity and reliability [43]. While focused on a specific application, this paper illustrates how invalid or manipulated data in IoT systems can lead to economic losses and system instability. This review discusses blockchain as a potential solution for data validation in IoT, highlighting the critical nature of this issue across various IoT applications [44].

Table 1 clearly shows how our proposed work addresses the limitations in existing research, particularly in terms of multi-fault detection, energy efficiency, and adaptation to remote environmental monitoring challenges. It provides a concise summary of the literature review and highlights the novel contributions of our research.

The number of wireless sensors used to collect data about the environment in outlying locations can vary from a few hundred to thousands. The data is delivered from the sensors to the central station, which in turn sends it to a server for analysis and forecasting. Internal or external causes could lead to data corruption [52]. Among these consequences are things like hardware or software problems, resource constraints, and the environment. Sensor nodes are used to collect data for both experimental and simulated purposes. As seen in Fig. 2, several sensors are linked to microcontrollers.

Sensors play a crucial role in collecting data from the surrounding environment. These devices are designed to capture various types of information such as temperature, humidity, and light level.

The collected data is typically intended to be utilized for monitoring, analysis, and decision-making processes. To ensure the reliability and accuracy of the collected data, state-of-the-art algorithms are employed for data validation [53]. These algorithms are designed to assess the integrity, consistency, and quality of the sensor data [54]. By employing sophisticated techniques, such as statistical analysis, and machine learning, the algorithms can identify and mitigate potential errors, outliers, or anomalies within the sensor readings. The data validation process involves comparing the sensor measurements against predefined criteria or expected patterns. This can include range checks, where the data is evaluated against predetermined minimum and maximum thresholds, or consistency checks, where the data is examined for logical coherence within the context of the monitored environment. For instance, if a temperature sensor placed indoors suddenly reports an extreme temperature reading, the validation algorithm can flag it as an outlier and discard or further investigate the data point [55].

Once the sensor data has undergone validation, it is ready to be transmitted to a server or a centralized system for further processing and analysis. The server can leverage this validated data to generate insights, detect patterns, trigger alerts, or facilitate decision-making processes. Additionally, the server can store the data for historical analysis, trend identification, or long-term monitoring purposes [56]. Overall, the utilization of sensors in conjunction with advanced data validation algorithms ensures the reliability, accuracy, and usefulness of the collected environmental data. By employing these professional techniques, organizations and systems can confidently rely on the sensed data to make informed decisions, optimize operations, and gain valuable insights into the environment being monitored. We developed a hybrid algorithm for identifying multiple data faults. The sample algorithm is shown below in Algorithm 1.

Justification for Choosing Artificial Neural Networks (ANN):

While more advanced techniques like deep learning models, ensemble methods, or advanced statistical approaches could potentially offer higher accuracy in some scenarios, we chose to use ANNs for our fault detection system for several reasons: 1.

Balance of performance and computational efficiency: ANNs offer a good balance between detection accuracy and computational complexity. In resource-constrained environments like remote sensor networks, this balance is crucial.

However, we acknowledge that this choice comes with limitations. More advanced techniques might offer higher accuracy in certain scenarios, and future work could involve comparing the performance of our ANN-based approach with other state-of-the-art methods like Random Forests, Support Vector Machines, or deep learning models.

Our choice of ANN represents a pragmatic approach that balances performance, efficiency, and implementation ability in the context of remote sensor networks. We believe it serves as a strong baseline for addressing the multi-fault detection problem while leaving room for future improvements and comparisons with more advanced techniques.

To ensure the accuracy of our algorithms, we chose to measure easily measurable characteristics such as temperature and humidity. We then constructed a testbed to record these readings and used MATLAB to run the models and collect data samples.

Data Collection and Preprocessing:

Sampling interval: Sensor measurements were collected at 5-minute intervals over 3 months

Dataset Preparation:

Training set: 70% of the data (18,396 samples)

ANN Configuration:

Architecture: 3-layer feedforward neural network

Fault Injection: To test the system’s fault detection capabilities, we artificially injected faults into the test set:

Spike faults: 5% of the data points

Performance Metrics: We evaluated the performance using the following metrics:

Accuracy: Percentage of correctly classified instances

Comparative Analysis: We compared our ANN-based approach with three existing methods:

Rule-based approach

The fault is out of range. Fig. 3 below shows the out-of-range fault. In this case, the original information is shown in yellow, and data that is outside the bounds is indicated with a red line.

By comparing the sensor data with the rest of the data, the modified z-score can identify outliers while ignoring stuck-at faults. Fig. 4 displays the results.

Whenever we feed data into our cutting-edge algorithm, it finds and properly identifies every single error. Fig. 5 displays the results.

To validate data autonomously in remote environmental monitoring, this study introduces an efficient and reliable IoT-based WSN architecture. The system solves important problems with data dependability and energy optimization by using an ANN-based fault detection method. We have made a vital addition to environmental monitoring systems with our solution, which outperformed existing approaches in terms of defect detection accuracy and energy economy.

A number of caveats need to be recognized notwithstanding these improvements. To start, in settings with limited resources, the AI-based fault detection algorithm’s computational complexity could lead to higher energy usage. Second, in some unusual or distant places, you might not have easy access to the high-quality training data that the algorithm needs to function properly. Finally, when the number of sensor nodes grows, the framework may encounter scalability issues, which could cause processing delays and extra communication costs. Furthermore, while the system performed well in simulations, limited real-world testing implies its adaptation to varied climatic circumstances remains questionable. Finally, the algorithm may require retraining or modification to handle novel defect types and may also be prone to false positives.

More extensive field testing, investigation of more dynamic network situations, and enhancements to the system’s scalability and adaptability are all in the works to remedy these shortcomings. By continuing to enhance the system, it is predicted that the proposed framework will become even more effective in facilitating dependable, energy-efficient data validation in challenging remote deployments.