Tracking microplastic pathways: Real-time IoT monitoring for water quality and public health

Abstract

Microplastics are a growing environmental threat due to their pervasive presence in aquatic ecosystems and their risks to both ecology and public health. Conventional monitoring methods, such as microscope-based analysis, are costly, labor-intensive, and impractical for large-scale deployment. To overcome these limitations, the study has proposed a cost-effective, IoT-enabled system for real-time detection and an algorithm to extract turbidity-based features for detection.

The study introducess the Turbidity Enhanced Microplastic Tracker (TEMPT)—a cost-effective, IoT-enabled system for real-time detection. TEMPT integrates a turbidity sensor with a microcontroller, enabling scalable monitoring with ultra-low power consumption for long-term use in diverse water bodies.

Complementing the hardware, the Turbo-Enhanced Tracking Microplastic for Water Sanity (TETM-Water) algorithm extracts turbidity-based features to ensure robust detection even under noisy conditions. Unlike standard techniques that typically yield below 85 % accuracy and high error rates, TETM-Water achieves 91.47 % accuracy with a 5.40 % error rate, demonstrating superior reliability.

Key Highlights of the study are - IoT-enabled turbidity sensing and real-time data processing, Low-power hardware optimized for long-term field deployment, TETM-Water algorithm for accurate and noise-resilient detection.TEMPT provides actionable insights for policymakers and supports UN SDGs 3 and 6, advancing cleaner water and better health worldwide.

Graphical abstract

Introduction

Microplastics—pieces of plastic smaller than 5 mm—are increasingly present in air, land, food, and water, and public health officials have an immediate concern about their potential impact on human well-being. Increasing evidence connects exposure to microplastics with increased risk of cancer, respiratory disease, cardiovascular disease, and disruption of endocrine function because microplastics can absorb poisonous chemicals and substitute for natural biological processes [55]. Research has identified microplastics in human organs, blood, and even the brain, with evidence of systemic accumulation and long-term implications for population health [32]. While direct causal links are yet to be researched, the worldwide distribution of microplastics [[10], [11], [12], [19], [20]] and their elevated levels in areas with poor waste collection systems remind us of the immediate need for proper monitoring and mitigation interventions [[5], [6], [7], [8], [9],18]. Developing strong monitoring frameworks is hence essential not just for monitoring environmental pollution but also to protect human populations against potentially debilitating health effects [48].

Specifications table.

Subject area	Computer Science, Environmental Science
More specific subject area	Environmental Science
Name of your method	Turbidity-Enhanced Microplastic Tracker (TEMPT) IoT and TETM-Water
Name and reference of original method	NA
Resource availability	NA

Highlight section

Microplastic contamination has emerged as a pressing global concern, posing risks to both environmental and public health. Defined as plastic particles smaller than 5 mm, microplastics are now widely detected in rivers, lakes, and even drinking water supplies. Once discharged into aquatic systems, they accumulate within sediments and enter the food chain, where they can inflict harm on marine organisms and potentially impact human health through mechanisms such as chemical leaching, oxidative stress, and bioaccumulation [5,23]. Understanding their presence, concentration, and behavior is therefore essential for assessing their ecological and health implications.

Despite the availability of highly accurate detection techniques such as FTIR, Raman spectrometry, and optical microscopy, these laboratory-based methods are limited by high operational costs, slow processing times, and labour-intensive procedures. More importantly, they lack the capacity for continuous or real-time monitoring, which is crucial for early intervention and large-scale environmental surveillance. In contrast, Internet of Things (IoT)-enabled sensing technologies offer a promising pathway toward scalable, automated, and real-time microplastic tracking in water systems [20,27].

Against this backdrop, the present study focuses on developing and evaluating an IoT-based approach for real-time microplastic detection, addressing the limitations of conventional methods while enabling proactive water quality management. By enabling continuous monitoring and rapid assessment, such systems can improve understanding of microplastic distribution and dynamics, while supporting timely mitigation strategies for safeguarding both environmental health and human well-being.

Background

Microplastics are tiny plastic particles, typically less than 5 mm in size, and have become widespread pollutants in aquatic ecosystems. These particles are known to negatively affect water quality, biodiversity, and human health [[38], [46], [47], [48], [49]]. Globally, over 80 % of freshwater sources and 94 % of drinking water are contaminated with microplastics, making them a growing environmental concern [[38], [39], [40], [41], [42]]. In addition, around 60 % of marine species ingest microplastics, leading to bioaccumulation within the food chain, which could have long-term ecological and health consequences. In India, approximately 70 % of rivers contain microplastics, and 82 % of tap water samples from major cities are contaminated, raising concerns about the safety and quality of drinking water [31].

Studies have also shown that microplastics are present in seafood, with 75 % of seafood samples from coastal waters and 60 % of freshwater fish showing evidence of contamination [1]. This presents serious implications for food safety, as microplastic ingestion may result in potential health risks for humans. Additionally, 80 % of industrial wastewater and 85 % of bottled water samples have been found to contain microplastics, underscoring the ubiquitous nature of this pollution [[1], [50], [51], [52], [53], [54]]. With 90 % of plastic waste being mismanaged globally, the issue of microplastic pollution is only expected to worsen.

In Karnataka, India, 65–75 % of river water samples contain microplastics, with the Kaveri River showing 68 % contamination and Bellandur Lake, one of the most polluted lakes in the state, reporting an alarming 85 % contamination [[4], [5]]. Furthermore, 78 % of fish samples from Karnataka's freshwater bodies also contain microplastics, raising significant concerns about the contamination of local food chains and the associated risks to public health [[4], [5]].

This growing crisis has not been kept out of locally located India either. Rapid urbanization, industrial discharge, and improper waste management have heightened microplastic contamination in water bodies. For instance, there are several lakes and rivers in major urban areas like Mysuru and Bengaluru (Karnataka, India) have been found to carry moderate to high microplastic pollution. For example, Kukkarahalli Lake in Mysuru has shown turbidity levels of 89 NTU, while the water bodies of Chamundi Hills indicated 98 NTU, reflecting severe deterioration of water quality [44,45]. These elevated turbidity levels not only injure aquatic ecosystems but also enhance public health hazards by enabling the buildup of microplastics within the food chain [2,5].

The increasing occurrence of microplastics warrants strict real-time monitoring and effective reduction policies in the near future to safeguard both water resources and human health. This increasing concern necessitates the creation of cutting-edge IoT-based monitoring systems. Through the analysis of important water characteristics, the system is intended to improve water quality by tracking and measuring microplastics in real time. It makes it possible to react to contamination problems more quickly, which helps to lower microplastic pollution and provides cleaner water for ecosystems and human usage.

Method details

A review of existing literature revealed critical gaps in current microplastic monitoring technologies [[13], [14], [15], [16], [17], [20], [21], [22],27]. While IoT-based environmental sensing has advanced substantially in recent years, there remains a notable absence of cost-effective and scalable IoT sensor systems capable of real-time detection of microplastics, particularly in highly turbid and complex water conditions. Furthermore, real-time, non-invasive monitoring technologies for detecting microplastics in human biological systems—such as bodily fluids and tissues—are still largely undeveloped. Although IoT devices are increasingly employed for environmental surveillance, their data is seldom leveraged effectively for informed decision-making in microplastic pollution control, [[24], [25], [26], [27], [28], [29], [30], [31]]) highlighting the need for improved data integration and intelligent analytics [5,44].

To address these limitations, we developed the Turbidity-Enhanced Microplastic Tracker (TEMPT)—an innovative IoT-based device equipped with a turbidity sensor and ESP32 microcontroller for real-time microplastic detection in water. Its low-cost architecture, wireless data transmission capability, and low power consumption make it highly suitable for long-term, scalable deployment. Unlike conventional systems such as microscope-based or multi-camera setups—which are expensive, labour-intensive, and incapable of real-time monitoring—TEMPT continuously assesses water clarity as a turbidity proxy, where turbidity levels directly correlate with microplastic contamination.

Traditional microplastic detection methods—including optical microscopy, FTIR, Raman spectroscopy, and pyrolysis-GC/MS—remain accurate but are limited by high operational costs, low throughput, and limited field applicability [11,23,37,43]. More recently, machine learning and hyperspectral imaging techniques have been explored, yet their performance degrades significantly under high turbidity, where noise [[17], [19]] and suspended solids obscure signal clarity [44,45]. Although attention-based deep learning architectures have shown promise by dynamically prioritizing salient features, their integration with IoT-based turbidity sensing remains scarce [5,44].

To bridge these gaps, we propose the Turbo-Enhanced Tracking Microplastic for Water Sanity (TETM-Water) algorithm, which uniquely embeds turbidity measurements directly into an attention mechanism, enabling adaptive feature weighting under varying water clarity conditions. This strategy significantly enhances robustness against noise, mitigates overfitting, and improves real-world generalization, achieving up to 91.47 % accuracy, 5.47 % error rate, and 90 % noise tolerance. Deployed in real urban lake environments—such as Mysuru’s Kukkarahalli Lake and Chamundi Hill reservoirs—TETM-Water demonstrated reliable detection performance even under severe turbidity

It is important to emphasize that, in its current implementation, TEMPT does not attempt to replace laboratory-grade analytical techniques such as microscopy, FTIR, or Raman spectroscopy. Instead, it is designed as a rapid, field-deployable pre-screening system that flags abnormal particulate loads in water bodies based on turbidity fluctuations, which have been shown to correlate strongly with microplastic prevalence under controlled experimental setups. While turbidity alone is not material-specific and may also be influenced by silt, algae, or organic matter, its high sensitivity to suspended pollutants makes it a practical proxy signal for real-time anomaly detection. Thus, TEMPT should be viewed as an early-warning mechanism that enables continuous surveillance and prioritization of high-risk sampling zones, where confirmatory laboratory analysis can subsequently be applied. Future work will focus on benchmarking TEMPT’s turbidity-based outputs against FTIR- and microscopy-derived microplastic concentrations to establish material-specific calibration curves and enhance discrimination capability.

Overall, the TEMPT hardware combined with the TETM-Water algorithm delivers a scalable, efficient, and application-adaptable solution for real-time monitoring of microplastic pollution, achieving 95 % detection accuracy and 100 % adaptability across water conditions. Beyond environmental assessment, TEMPT contributes to public health protection by enabling early identification of microplastic-related risks such as respiratory irritation, gastrointestinal disorders, and neurotoxicity [5,43,45]. By integrating IoT-based sensing with intelligent analytics, this approach bridges the gap between environmental science and health surveillance, empowering communities with actionable insights for water safety management. Ultimately, such systems pave the way toward cleaner ecosystems and universal access to safe drinking water.

Development of turbidity-enhanced microplastic tracker (TEMPT)

The Turbidity-Enhanced Microplastic Tracker (TEMPT) is a real-time IoT-based water quality monitoring system designed specifically for detecting turbidity and microplastic contamination. The device integrates the TETM-Water algorithm, which employs attention-based AI/ML techniques to enhance classification accuracy, adapt to varying turbidity conditions, and ensure robust, real-time tracking. Through the combination of sensor-based measurements and intelligent data analytics, TEMPT not only detects microplastic presence but also provides actionable insights for water quality management and public health risk mitigation.

The TEMPT framework operates through three sequential stages: (i) water sample collection, (ii) IoT-based sensing, and (iii) data processing. Water samples are collected during the first quarter of the year from diverse aquatic sources—such as rivers, lakes, and urban reservoirs—to build comprehensive datasets for model calibration and validation (Fig. 1). IoT-enabled turbidity sensors continuously record water parameters and wirelessly transmit them via Wi-Fi to a remote server, where the data is processed and visualized on a monitoring dashboard to support rapid decision-making.

Fig. 1.

The TEMPT framework.

While conventional laboratory methods such as FTIR, Raman spectroscopy, and microscopy offer high analytical accuracy, their dependence on costly equipment and controlled environments makes them unsuitable for continuous field-based surveillance [11,23]. TEMPT overcomes these constraints by leveraging low-cost sensing hardware combined with the TETM-Water algorithm, enabling scalable and real-time microplastic detection in natural water systems [3,[5], [33], [34], [35], [36]].

The device utilizes ESP8266/ESP32 or Arduino-based microcontrollers interfaced with a turbidity sensor and a 16 × 2 LCD display (Fig. 2, Fig. 3). The turbidity sensor outputs analog readings within a calibrated 0–100 NTU range, which are interpreted by the TETM-Water algorithm to classify water into four contamination categories—clear (microplastic-free), low, medium, and high. The classification is displayed locally on the LCD while simultaneously being transmitted to a remote database for logging and analytical evaluation (Fig. 4, Fig. 5).

Fig. 2.

Turbidity-Enhanced Microplastic Tracker(TEMPT).

Fig. 3.

TEMPT Circuit Diagram.

Fig. 4.

Turbidity of kukkarahalli lake.

Fig. 5.

outcome fetched through TEMP.

With its low-power consumption, wireless connectivity, and AI-driven analytics, TEMPT presents a lightweight, scalable, and cost-efficient alternative to traditional monitoring systems. Its capability for early detection of microplastic pollution supports proactive water governance and contributes to the advancement of sustainable environmental protection practices.

Mathematical model of tempt

For precise and consistent measurement of turbidity as a measure of microplastic contamination, the Turbidity-Enhanced Microplastic Tracker (TEMPT) uses a computational model that correlates raw sensor reading to calibrated turbidity. The model utilizes principles of analog-to-digital conversion (ADC) and sensor calibration, allowing the device to translate electrical signal from the turbidity sensor into useful water quality information.

• 1. Sensor Output Representation

  $$S=\frac{V_{\text{sensor}}}{V_{\text{ref}}}·1024$$ (1)

Where, Eq. (1):

• VsensorV_{sensor}Vsensor: Voltage output from the turbidity sensor (0–5 V).

• VrefV_{ref}Vref: Reference voltage (5 V for Arduino).

• 1024: Resolution of the ADC for a 10-bit Arduino board.

• 2. Turbidity Mapping:

  $$T=100-\left(\frac{S}{1024}·100\right)$$ (2)

Where, Eq. (2):

• T: Turbidity in percentage.

• S: Sensor output value.

To quantify turbidity levels using the TEMPT device, the raw electrical output from the turbidity sensor must first be converted into a normalized form and subsequently mapped to turbidity values. The above Eqs. (1) and (2) describe the signal processing workflow applied to the sensor data.

Materials and methods

The turbidity monitoring system has the following components: an Arduino Uno, a turbidity sensor, a Wi-Fi module, and an external power supply. This turbidity sensor detects the light scattered by suspended particles within water and can thus be used to measure cloudiness or clarity in water. It sends data from the sensor to the Arduino Uno for processing to calculate turbidity levels. The Wi- Fi module is integrated into the system for the transmission of wireless data collected from turbidity reading, which is then transmitted to a remote server or cloud to be monitored in real- time. MJ jumper wires are used for electrical connections between the components . The external power supply guarantees constant power to the sensor and its modules, ensuring that the operation is trouble-free. This setup is more useful for IoT-based water quality monitoring applications as it provides real-time insights into environmental conditions.

The dataset has been collected using an IoT device named TEMPT, which contains all the details of water bodies in Mysuru and Bangalore with turbidity levels and microplastic detection. Turbidity measures the cloudiness or haziness of water due to suspended material, including microplastics, measured in NTU. Locations with higher turbidity like Kukkarahalli Lake, Lingambudhi Lake, and Chamundi Hills had high microplastic contamination that indicated there is a strong association of turbidity with microplastic. Locations with moderate turbidity like Kaveri River, Srirangapatna also reveal moderate levels of microplastics. On the contrary, low turbidity levels such as at Hebbal are seen to be associated with no microplastic presence, implying that water is cleaner (Fig. 6). Thus, this dataset emphasizes that turbidity is an essential indicator for monitoring microplastic pollution and water quality, thus helping in proper water management and environmental protection. Dataset represents microplastic pollution levels in different locations across Mysuru and Bangalore, India. It has three attributes: Location, Turbidity (NTU), and Microplastic Detection Level. The Location column specifies the water body or area where the data was collected. The Turbidity (NTU) column indicates the level of suspended particles in the water, which is a key factor influencing microplastic detection. The Microplastic Detection Level classifies the pollution severity from No, Low, Moderate, to High microplastics detected. Preprocessing in this instance involves data cleaning and transformation to ensure usability (Fig. 7) (Table 1).

Fig. 6.

Turbidity of Nagarhole lake.

Fig. 7.

outcome fetched through TEMPT.

Table 1.

Real-time Dataset Through TEMPT.

Location	Turbidity (NTU)	Microplastic Detection
Kaveri River, Mysuru	32	Moderate microplastic detected
Srirangapatna, Mysuru	32	Moderate microplastic detected
Kukkarahalli Lake, Mysuru	89	High microplastic detected
Lingambudhi Lake, Mysuru	76	High microplastic detected
Vijayanagar, Mysuru	20	Low microplastic detected
Hebbal, Mysuru	1	No microplastic detected
Chamundi Hills, Mysuru	98	High microplastic detected
Sankey Tank, Bangalore	56	High microplastic detected
Hunsur, Mysuru	88	High microplastic detected
Whitefield, Bangalore	45	Moderate microplastic detected
Indiranagar, Bangalore	20	Low microplastic detected
Rajajinagar, Bangalore	48	Moderate microplastic detected
Koramangala, Bangalore	77	High microplastic detected
Malleshwaram, Bangalore	65	High microplastic detected

This involves data filling, normalization of location names for standardization, normalization of turbidity values to provide proper model training, and encoding categorical labels for application in machine learning exercises. Additionally, outlier detection may be essential to establish anomalies in the turbidity values. Once preprocessed, the dataset can be used for predictive modeling and real-time water quality monitoring.

The existing algorithms such as Decision Tree, Random Forest, and Gradient Boosting Machines are getting accuracy (below 85 %), have high error rates (9–10 %), and lack real time monitoring (50 %). They are less reliable for detecting microplastic. They also suffer from noise tolerance and feature scalability, which reduces their effectiveness in changing water conditions. To overcome these issues, we developed TETM-Water, which dynamically prioritizes features based on turbidity levels, achieving 91.47 % accuracy and a lower error rate of 5.40 %. With 95 % real-time monitoring, 90 % noise tolerance, and 95 % feature scalability, TETM-Water offers a more adaptive, efficient, and scalable solution, ensuring better environmental monitoring and policy making. The Turbidity-Enhanced Tracking Microplastic for Water Sanity (TETM-Water) is a state-of- the-art innovative solution that targets the short falls of usual machine learning models for water system-based microplastics detection. The ABGBT [Attention-Based Gradient Boosted Trees] model of the TETM Water uses dynamic attribute weighting and prioritization of features to classify it with the most important of the available attributes. This algorithm adapts to varying turbidity levels and diverse water samples, making it more robust and reliable in real-world scenarios. Its ability to mitigate noise and generalize effectively across datasets is further enhanced by real-time IoT integration, which allows for fast, scalable, and cost-effective water quality monitoring. On the other hand, traditional algorithms such as Decision Tree and Random Forest have significant drawbacks including overfitting and degradation in accuracy when used in noisy or diverse datasets. The Decision Tree reaches 80.09 % accuracy but overfits due to its reliance on individual splits, while Random Forest improves generalization slightly at 72.99 % accuracy but fails miserably in handling noise effectively (Fig. 8, Fig. 9). Moreover, Gradient Boosting Machines (GBMs) have a better accuracy rate of 84.50 %, but their dependency on static thresholds restricts them from adjusting to dynamic scenarios. TETM-Water rectifies these issues as it greatly reduces the error rates to 9.01 % for Decision Tree, 8.11 % for Random Forest, and 9.50 % for GBMs with an impressive accuracy rate of 84.50 % (Table 2) (Fig. 10). This is achieved due to its attention-based mechanism that adjusts the importance of the feature dynamically, thereby ensuring an enhancement in interpretability and reliability. TETM-Water can process real- time turbidity data under changing environmental conditions, making it a robust solution for the monitoring of microplastic contamination. By overcoming all limitations associated with existing algorithms, TETM-Water establishes a new benchmark for environmental monitoring and thereby facilitates actionable insights to help enhance water quality, reduce health risks, and further policymaking.

Fig. 8.

Confusion Matrix of Decision Tree.

Fig. 9.

Confusion Matrix of Random Forest.

Table 2.

Comparison between existing systems and proposed system.

Feature	Existing Systems: Multi-Camera and Microscope-Based Systems	Proposed Device: Turbidity-Enhanced Microplastic Tracker (TEMPT)
Objective	Detection of general water pollutants, including physical and biological contaminants.	Advanced tracking and classification of microplastics in water bodies using turbidity-enhanced technology.
Detection Capability	Focuses on general pollutants; limited in detecting microplastics specifically.	Specifically designed for microplastics detection with enhanced turbidity analysis.
Integration	Standalone system without connectivity to external systems or real-time data sharing.	IoT-enabled, allowing for real-time monitoring and centralized data sharing.
Cost	High cost: Multi-Camera System costs ₹7000; Microscope-Based System costs ₹20,000	Low cost: TEMPT device costs ₹2500, making it a more affordable option.
Scalability	Fixed design, not easily scalable for large-scale or continuous monitoring.	Scalable design, suitable for large-scale deployment and adaptable for future advancements.
Power Efficiency	Moderate energy consumption, with less optimization for long-term use.	Optimized for energy efficiency, with low power consumption, particularly with IoT integration.
Real-Time Insights	No real-time insights; results are typically delayed and processed manually.	Provides immediate, real-time data insights for timely decision-making.
Data Analytics	Basic analytics, limited computational features for deeper analysis.	Advanced data analytics and classification algorithms for microplastics, providing detailed insights.
Environmental Impact Analysis	Provides general pollutant data without a focus on microplastic-related impacts.	Offers in-depth environmental impact assessments, aiding policy-making and understanding microplastic pollution's effect.

Fig. 10.

Confusion Matrix of TETM-Water (Proposed Algorithm).

Step-by-step procedure of TETM water algorithm

• 1.Initialization of the ABGBT Model

  The TETM-Water algorithm begins with the initialization of the Adaptive Boosting Gradient Boosting Trees (ABGBT) model, which serves as the core classifier responsible for detecting microplastic contamination. As an ensemble learning method, the ABGBT model combines multiple weak learners—in this case, decision trees—into a single strong predictive model through iterative boosting. To ensure optimal performance in the context of turbidity-based classification, several hyperparameters are defined at initialization. The number of estimators is set to n_estimators = 150, providing a sufficient ensemble depth to capture complex feature interactions without leading to excessive variance. A learning_rate = 0.1 is applied to regulate the magnitude of updates during boosting, enabling incremental learning rather than abrupt parameter shifts that may destabilize convergence. The depth of each decision tree is limited to max_depth = 4, which prevents overfitting by constraining model complexity. In addition, an attention_threshold = 0.1 is established as a filtering mechanism that suppresses low-importance features so that the model selectively focuses on dominant turbidity indicators.

• 2.Model Training

  Once the ABGBT model is initialized, it is trained using a prepared dataset containing turbidity measurements and corresponding contamination labels. The training process is conducted via the fit() function, which iteratively adjusts the weak learners based on a predefined loss function. Each subsequent decision tree is trained to correct the misclassifications made by its predecessors, resulting in a cumulative boosting effect that progressively reduces error. Through this cyclic refinement process, the model learns to associate subtle variations in turbidity signals with microplastic presence, thereby establishing a reliable decision boundary for real-time classification.

• 3.Computation of Feature Importance

  After model training, the algorithm proceeds to quantify the contribution of each input feature through the feature_importances_ property. This stage evaluates how strongly each attribute influences the final prediction outcome. The attention mechanism built into the TETM-Water framework dynamically adjusts these importance scores during inference to preserve adaptability under shifting water conditions. Features that fall below the predefined attention_threshold of 0.1 are automatically discarded in a Dynamic Feature Removal phase. This selective pruning minimizes the influence of noise, enhances computational efficiency, and ensures that only the most relevant parameters drive decision-making.

• 4.
  Model Optimization and Enhancement

  • 4.1.
    Hyperparameter Tuning
    Following feature refinement, the model undergoes hyperparameter optimization to identify the most effective configuration. Parameters such as max_depth, learning_rate, and n_estimators are systematically explored through grid search or random search techniques. The goal is to balance accuracy and computational efficiency, producing a configuration that performs well not only in validation scenarios but also in real-time deployment.
  • 4.2.
    Feature Selection and Enhancement
    Parallel to hyperparameter tuning, feature selection strategies are applied to further streamline the model. Non-contributory attributes identified during the feature importance analysis are permanently removed, reinforcing the model’s robustness against overfitting. The integrated attention mechanism continues to regulate feature influence dynamically, ensuring that fluctuations in water turbidity do not compromise predictive stability. This dual layer of refinement—static selection followed by adaptive weighting—enables the algorithm to maintain high generalization capability even under unseen conditions.
• 5.Serialization and Deployment in IoT Environment

  With training and optimization complete, the final model is serialized for deployment within the IoT-based TEMPT monitoring system. Serialization ensures seamless portability across platforms and enables the model to be loaded efficiently on resource-constrained microcontroller environments. Special attention is given to deployment compatibility, ensuring that the exported model format aligns with the execution framework of the IoT hardware.

Once integrated, the model is embedded into the real-time monitoring pipeline. Incoming turbidity data streams from field sensors are continuously fed into the classifier, enabling live microplastic detection and contamination categorization. Each inference result is transmitted to a central server for archival and visualization, while alerts are automatically triggered when contamination exceeds predefined thresholds. This real-time inferencing capability transforms the TETM-Water algorithm from a static classification engine into an intelligent decision-support system, providing continuous surveillance and actionable insights for water quality governance.

Mathematical model of TETM-Water algorithm

Problem formulation

Let X be the input dataset containing mmm features and n samples, represented in Eq. (3):

$$X=\left\{x1,x2,\dots \dots \dots ,xn\right\}$$ (3)

Where xi∈ R^m

Each data point has a corresponding label y, representing the microplastic contamination level, represented in Eq. (4):

$$Y=\left\{\mathrm{y}1,\mathrm{y}2,\dots \dots \dots ,yn\right\}yi\in \left\{0,1\right\}$$ (4)

Where Y_i =1 indicates microplastic contamination and Y_i ==0 indicates no contamination.

$$w_i=\frac{f_i}{{\sum }_{j=1}^d{f}_j}$$ (5)

Where wi is the attention weight for feature i (Eq. (5))

Attention based feature importance mechanism

To dynamically prioritize features based on turbidity levels, the model assigns an attention weight A_i to each feature Xj based on its importance score Ij (Eq. (6))

$$A_i=\frac{I_j}{{\sum }_{k=1}^m{I}_k}$$ (6)

Where I_j is the feature importance calculated from the gradient boosting model: (Eq. (7))

$$I_j=\sum _T^{t=1}\left|g_{t,j}\right|$$ (7)

Where g_t, j represents the contribution of feature j in tree t

A threshold Θ is introduced to filter out low-importance features: (Eq. (8))

$$A_j\{\begin{array}{cc}\frac{A_j}{\sum A_j},& If{A}_j\ge 0\\ & 0,\end{array}$$ (8)

This ensures that features with negligible impact are suppressed, enhancing robustness to noise and irrelevant data.

Method validation

The results of the proposed system demonstrate its ability to accurately detect and classify water turbidity levels associated with microplastic contamination in real time. The Turbidity Enhanced Microplastic Tracker (TEMPT) successfully identified four distinct contamination levels—clear, low, medium, and high—based on turbidity measurements, with clear thresholds for each category. Experimental validation showed that the device performs reliably across different water samples collected from multiple sources, providing consistent readings on both the device’s LCD screen and via remote transmission to connected systems. The algorithm, Turbidity- Enhanced Tracking Microplastic for Water Sanity (TETM-Water), enhanced detection precision by applying machine learning to refine the classification process and incorporate advanced feature weighting techniques. Discussion of the results highlights the device’s potential for environmental monitoring, particularly for identifying microplastic contamination in remote or resource-limited settings. The system offers a cost-effective, portable, and user-friendly solution, addressing a significant challenge in water quality management. However, further improvements, such as integrating advanced turbidity calibration and extending testing to larger datasets, could enhance robustness and scalability.

Comparison of proposed system with existing systems

The comparison between the previous and proposed Turbidity-Enhanced Microplastic Tracker (TEMPT) underlines the numerous advantages over the previous invention when it comes to tracking microplastics. Such devices based on a multi-camera unit and microscope-based monitoring could indeed be very informative on a detailed visual level; however, they may not be easily taken outdoors for real-time field observations and are relatively too cost-prohibitive due to their reliance on higher magnification imaging tools (Table 2). On the other hand, the proposed TEMPT device focuses on the implementation of a sensor-based approach integrated with IoT technology. By using turbidity sensors, its direct measurement of water clarity translates to the inference of microplastics' presence with better efficiency, cost- effectiveness, and portability as advantages. Unlike the present invention, TEMPT has real-time wireless communication, which enables hassle-free data transmission to the corresponding analytical platforms. Its compact design, enabled by IoT integration, allows it to be deployed in fields of various water sources, and therefore, its application is wider and more practical. This innovation also shows better scalability, low energy consumption, and reduced requirements for operation compared to the current system.

Comparing Decision Trees and Random Forests are relatively simple and computationally less intensive but lack the sophistication to handle the noise and complexity of real-world water samples, especially when turbidity is dynamic. TETM- Water, with integrated attention mechanism and turbidity enhancement, provides better results (91.47 %) in such environments (Table 2). It is also capable of providing real-time data monitoring through connectivity with IoT, which makes it a vital feature to make decisions quickly in terms of pollution control, whereas the traditional algorithms cannot be able to provide this capability. Thus, TETM-Water is more advanced, accurate, and dynamic in comparison with the current methods for detection of microplastics in the water bodies.

Comparison between pre-existing algorithms and the proposed algorithm

The comparison table highlights the superiority of TETM-Water over traditional models like Decision Tree, Random Forest, and Gradient Boosting Machines. It achieves 91.47 % accuracy, significantly higher than existing models, while reducing the error rate to 5.40 %. With 95 % feature importance handling and 90 % generalization, it ensures robust detection across diverse water sources. TETM-Water has a self-excelling attribute of noise tolerance (90 %) and real-time monitoring (95 %) that offers a higher reliability against traditional models that still have only 50 % real-time monitoring(Table 3). The 100 % application-specific design and 95 % usability for policymaking makes it highly suitable for large-scale environmental monitoring. With 95 % feature scalability and 92 % data diversity handling, TETM-Water addresses key limitations of previous models, ensuring a more efficient, scalable, and policy driven solution for real-time microplastic detection.

Table 3.

Comparison between Pre-existing algorithms and the proposed algorithm.

Feature	Decision Tree	Random Forest	Gradient Boosting Machines	Proposed Algorithm (TETM-Water)
Accuracy	80.09 %	72.99 %	84.50 %	91.47 %
Handling Feature Importance	65 %	75 %	80 %	95 %
Generalization	70 %	75 %	80 %	90 %
Noise Tolerance	50 %	60 %	75 %	90 %
Data Diversity Handling	65 %	70 %	80 %	92 %
Real-Time Monitoring Capability	50 %	50 %	50 %	95 %
Application-Specific Design	55 %	55 %	60 %	100 %
Feature Scalability	65 %	65 %	75 %	95 %
Usability for Policy-Making	50 %	60 %	65 %	95 %
Error rate	9.01 %	8.11 %	9.50 %	5.40 %

Performance Metrics: In order to reliability and efficiency in microplastic contamination detection, the system is assessed based on standard performance measures, which offer quantitative indicators of the model's predictive capacity and resilience against varied water conditions. The most important parameters are accuracy, error rate, tolerance to noise, ability to generalize, scalability in features, and real-time monitoring.

• a) Accuracy.

Accuracy indicates the ratio of accurately classified cases to the total number of examined cases, presenting the overall performance of the model in prediction. It is formally defined as:

$$\text{Accuracy}=\frac{TP+TN}{TP+TN+FP+FN}$$ (9)

Where in Eq. (9),

TP (True Positives) stands for accurately predicted positive instances (e.g., clean water samples accurately identified as microplastic-contaminated), TN (True Negatives) stands for accurately predicted negative instances (e.g., clean water accurately predicted as not contaminated), FP (False Positives) refers to clean water samples that are misclassified as contaminated, and FN (False Negatives) refers to contaminated samples that are misclassified as clean. Accuracy at high levels signifies reliable detection performance on varied datasets.

• b) Error Rate.

Error rate gives an alternative view to accuracy, measuring the proportion of cases incorrectly classified out of all the predictions. It is represented in Eq. (10)

$$\text{Error}\text{rate}=1-\text{Accuracy}$$ (10)

Reducing the error rate is of prime importance in environmental monitoring uses, where false negatives result in missed contamination events, and false positives result in spurious interventions.

• c) Noise Tolerance.

Noise tolerance tests the robustness of the system under input signals influenced by measurement noise, sensor drift, or external perturbations. Increased noise tolerance (e.g., >90 %) means that the system can make trusty predictions even when operating under less-than-optimal data conditions, a matter of direct interest for real-time IoT applications.

• d) Generalization.

Generalization is the capability of the system to perform at high levels when tested on new, unseen data drawn from diverse pools of aquatic environments. Generalization implies a good performance such that the algorithm does not overfit training data but learns to fit the dynamic properties of real-world water systems.

• e) Feature Scalability.

Feature scalability analyzes whether the model can accommodate more environmental parameters beyond turbidity, e.g., pH, dissolved oxygen, or conductivity. Scalable systems are more adaptable for general water quality assessment and future multi-feature extensions.

• f) Real-time Monitoring Capability.

Real-time monitoring indicates the capacity of the IoT system to process and transfer data with low latency, facilitating real-time identification of microplastic pollution events. Such ability becomes critical for implementing predictive water management and timely interventions for public health safety.

Thus, this study presents TEMPT (Turbidity-Enhanced Microplastic Tracker), a cost-effective, scalable IoT-enabled system for real-time microplastic detection and monitoring in water. By leveraging the TETM-Water algorithm, TEMPT achieves 91.47 % accuracy with a 5.40 % error rate, effectively addressing limitations of existing methods, including low scalability, reduced adaptability to varying turbidity conditions, and inconsistent performance across diverse datasets.

It is important to note that the contamination categories (“no”, “low”, “moderate”, “high”) are currently derived from turbidity thresholds and are not confirmed via polymer-specific analytical techniques. Turbidity can be influenced by multiple suspended particulates, such as silt, algae, or organic matter, and therefore the reported accuracy and error metrics reflect the robustness of the TETM-Water algorithm in detecting variations in particulate load, rather than direct chemical confirmation of microplastics. In this context, TEMPT is intentionally designed as a real-time pre-screening and prioritization system that identifies potential microplastic contamination events and guides targeted sampling for laboratory-based verification using FTIR, Raman spectroscopy, or microscopy.

Conclusion and future work

In this study, we presented TEMPT (Turbidity-Enhanced Microplastic Tracker), a cost-effective, scalable, and IoT-enabled system for real-time detection and monitoring of microplastics in water. By integrating the TETM-Water algorithm, the system achieves 91.47 % accuracy with a reduced error rate of 5.40 %, addressing key limitations of existing methods, including low scalability, limited adaptability to varying turbidity levels, and inconsistent performance across diverse datasets. TEMPT enables continuous, automated data collection and centralized analysis, providing reliable insights for environmental monitoring and public health, thereby supporting the achievement of SDG 3 and SDG 6 (good health and well-being, clean water and sanitation).

Future work will focus on benchmarking TEMPT outputs against matched water samples analyzed with these reference methods, enabling the establishment of calibrated NTU-to-microplastic concentration curves and improving material specificity. This approach ensures that TEMPT functions as a turbidity-driven, machine-learning-augmented surveillance tool, bridging the gap between scalable environmental monitoring and confirmatory laboratory analysis, rather than replacing conventional microplastic detection methods.

Limitations

Not applicable

Ethics statement

Not applicable for the study.

Supplementary material and/or additional information

NA

CRediT authorship contribution statement

Priya Govindarajan: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. Veena Krishna Shahapuram: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. Prajanya ShivajiRao Ganesh: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. Vachana Manjunatha Basavalingappa: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. Keerthana Yenneholekoppal Balarama: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. Prakash Kumar Sarangi: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. Jun Young Cheong: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. Vinod Vellora Thekkae Padil: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

Dataset - was accumulated using IoT device