Comprehensive Study of Statistical, Machine Learning, and Heuristic Approaches of Anomaly Detection Techniques in Wireless Sensor Networks : Research Article

Comprehensive Study of Statistical, Machine Learning, and Heuristic Approaches of Anomaly Detection Techniques in Wireless Sensor Networks : Research Article

Geeta

Research Scholar, Department of Computer Science and Engineering, Jagannath University Bahadurgarh (Delhi NCR), India.

Dr. Renuka Arora,

Professor, Department of Computer Science and Engineering, Jagannath University Bahadurgarh (Delhi NCR), India

Abstract – Due to the proliferation of WSNs in real-time monitoring and control environment, anomaly detection in WSNs has become a significant area of research. These networks are particularly wired to data corruption due to faulty sensors, environmental inferences, malicious attacks, lack of communication or breakdowns. A comprehensive survey of different types of anomaly detection algorithms in WSNs, including statistical and machine learning-based (e.g., deep learning and rule-based). The ablation study delves into super- vised, unsupervised and semi-supervised models, such as SVM, DT, clustering, autoencoders, and LSTM-based architectures. Special emphasis is devoted to lightweight and energy-aware algorithms aimed at WSN nodes with limited resources. Further, this paper considers challenges of the malware hotspot including high false-positive rates, dynamic characteristics of network topologies, data imbalance, and scalability. Simulated experiments on MATLAB illustrate the effectiveness and trade- offs of the proposed methods. The findings highlight the significance of hybrid and context-aware systems for effective and real-time anomaly detection in WSN environment with heterogeneity and low power.

Keywords – Wireless Sensor Networks (WSNs), Anomaly Detection, Machine Learning, Deep Learning, Rule-Based Systems.

1. INTRODUCTION

   Wireless Sensor Networks (WSNs) Wireless Sensor Network (WSN) is an ad-hoc network technology applied to a constrained environment, formed by spatially distributed autonomous devices using sensors to monitor physical or environmental conditions, such as temperature, sound, vibration, pressure, motion or pollutants. These nodes work together to relay gathered data to a central base station (or sink node) for processing. Because of their flexibility and scalability with relatively low-cost deployment, WSNs are commonly applied in the areas of environmental observation,

   smart agriculture, military surveillance, health-care system, and industrial automation.

   Importance of Data Integrity and Security: Although these advantages, there are a lot of challenges for WSNs, especially for maintaining data consistency and achieving secure communications. The information sensed by WSN nodes is usually relevant and time-critical. If this data becomes corrupted or compromised in any way, then potentially serious decisions could be made in a wrong way, which could also be life threatening in mission-critical applications such as battle field monitoring, or disaster management. Due to the modest computational power, energy constraints and wireless nature, these networks are especially prone to anomalies and security threats.

   Anomalies in WSNs: Anomalies in WSNs denotes the unusual or unexpected trends in sensor data or the network operations other than normal trends. These anomalies can be triggered by different causes for example hardware malfunctioning (e.g. sensor failure), environmental changes (e.g., temperature spiking), or even attacks (e.g., Sybil or sinkhole). Timely and precise anomaly detection is crucial if WSN applications are to be reliable and robust.

   Objectives of the Study. The purpose of this paper is to:

   • Provide a comprehensive overview of anomaly types in WSNs.

   • Review and compare existing anomaly detection techniques, including machine learning and statistical approaches.

   • Highlight current challenges and limitations in the field.

     A. Anomalies in WSNs

     In a WSN, an anomaly is defined as any unexpected phenomenon or abnormal pattern which is different from the normal behavior of network. These anomalies can be

     triggered by hard-ware anomalies, environmental disturbances, or malicious attacks. Although a few misuses in the network are likely to go undetected, anomalies are threats to the accuracy, efficiency, and security of the network, thus, it is critical to diagnose them to maintain the network reliability. The major anomalies detected in WSNs are:

     Node Failure: Cyprus anomalies: node failures are the major reasons of WSNs, due to the rough operating conditions, limited battery capacity and hardware vulnerability of sensor nodes. Failures can cause abrupt data loss, circuit desaturation, and incomplete data transmission.

     Environmental Anomalies: Sensor nodes could record data that is substantially outside the expected spectrum because of sudden shift in environmental conditions. These can be sudden temperature extremes, abnormal chemical concentrations, or objects in the path.

     Malicious Behavior or Security Attacks

     Given their decentralized and wireless nature, WSNs are vulnerable to various security threats. Malicious anomalies can arise from:

   • Sybil attacks, where a single node presents multiple identities to disrupt routing or voting mechanisms.

   • Sinkhole attacks, where a compromised node attracts traffic and drops or alters packets.

   • Selective forwarding, where an attacker node forwards only selected packets and drops others.

   • False data injection, where fake readings are sent to manipulate the monitoring outcomes.

     These attacks not only corrupt the data but may also affect routing, synchronization, and network availability.

     Impact of Anomalies on Network Performance

     The presence of anomalies in WSNs can severely degrade network performance across several dimensions:

   • Data Integrity: Inaccurate or falsified sensor data leads to incorrect decision-making, especially in mission- critical systems like disaster response or healthcare monitoring.

   • Energy Efficiency: Anomalies such as repeated retransmissions due to faults or malicious flooding attacks result in unnecessary energy consumption, shortening node and network lifetime.

   • Network Throughput and Latency: Anomalies may introduce delays, congestion, or loss of communication paths, affecting real-time data delivery.

   • Topology Changes and Routing Instability: Node failures or attacks can alter the routing paths and network topology, increase maintenance overhead and reduce reliability.

   • Security Breaches: Malicious activities compromise the confidentiality, integrity, and availability of the network, leading to potential data leaks or system failures.

   Early and accurate detection of these anomalies is therefore essential for ensuring the robustness and sustainability of WSNs. The next section provides a detailed exploration of techniques developed for anomaly detection.

2. LITERATURE REVIEW

   TABLE 1. RECENT MACHINE LEARNING AND AI-BASED ANOMALY DETECTION APPROACHES IN WIRELESS SENSOR NETWORKS

   S. No.	Author(s)	Proposed Work	Methodology	Conclusion
   1	Revanesh et al. (2024)	ANN-based LMN for energy efficiency and anomaly detection	Integration of refined LMNN with LEACH and EESR routing protocols	Achieved superior performance in energy efficiency and anomaly detection accuracy
   2	CL and Sumathi (2024)	Consensus-based anomaly detection (CSAD)	Three-step approach: classification, level- based decision, and elimination	CSAD outperformed baseline with higher throughput (81.81%)
   3	Gayathri and Surendran (2024)	Unified anomaly detection via EFL and OAD-EE	Combined federated learning and online anomaly detection with energy-efficient techniques	Achieved highest detection accuracy (EFL) and lowest energy use (OAD-EE)
   4	Sharma et al. (2024)	ML-based anomaly detection with energy optimization	Region-based clustering and anomaly detection using RF, KNN, and LOF	KNN improved stability period by 259%, RF by 246.4%, outperforming traditional methods
   5	Wei (2024)	AI-based anomaly detection for cardiopulmonary monitoring	Integrated learning algorithm including traffic analysis and feature extraction	Outperformed traditional methods in detection efficiency and response time
   6	Ahmad and Alkhammash (2024)	Online Adaptive Kalman Filtering (OAKF) for real-time detection	Adaptive filtering with live threshold tuning	Achieved 95.4% accuracy, 0.008s processing time, suitable for real-time WSNs
   7	Xin et al. (2024)	Survey on ML techniques for anomaly detection	Evaluation of supervised, unsupervised, and semi-supervised learning	Highlighted strengths and limitations, offering future directions
   8	Mohan et al. (2024)	EdgeAnomaly using GANs at the edge	Training GANs on normal data for real- time anomaly detection	Demonstrated strong detection in dynamic edge WSNs
   9	Haque et al. (2023)	ML-based anomaly detection survey in WSNs	Categorization of ML techniques and comparative analysis	Outlined challenges and potential research areas in ML-based detection
   10	Chen et al. (2023)	BS-iForest for improved anomaly detection	Box-plot-based sampling and high- accuracy isolation trees	Improved AUC by 1.5%7.7% over standard iForest
   11	Ravindra et al. (2023)	ETELMAD model using optimized ELM	Data compression, ELM prediction, dynamic thresholding	Achieved 97.4% accuracy on IBRL dataset
   12	Kumar et al. (2023)	STA-Tran: Transformer-based anomaly detection	Spatio-temporal attention with transformer architecture	Achieved F1 score of 0.9109, precision 0.9191, high detection accuracy

   13	Arul and Venkatesan (2023)	FANN for anomaly detection in WSNs	Feed-forward autoencoder for data reliability	Reduced false alarms, enhanced energy efficiency
   14	Srivastava and Bharti (2023)	HMOI: Hybrid of One-Class SVM and Isolation Forest	Conversion of unlabeled data to labeled, followed by detection	Accuracy >99%, false alarm rate 0.08%
   15	Premkumar et al. (2023)	SEECAD: DoS detection in WSNs	Cluster-based detection without cryptography	Enhanced detection and network lifetime
   16	Mittal et al. (2022)	QoS-oriented WSN with ML	Used LEACH, Sub-LEACH, LMNN, and MFO; applied ML classifiers	SVM achieved highest detection; Sub- LEACH + MFO performed best
   17	Yao et al. (2022)	PCA + DCNN for DoS detection	Lightweight deep learning model for anomaly detection	Outperformed traditional models with improved classification
   18	Zhang et al. (2022)	Dynamic GNN-based anomaly detection	Integrated temporal, spatial, and modal features	Achieved F1 score of 0.90, 14.2% above baseline
   19	Wang et al. (2022)	Adaptive sliding window for time- series detection	Trend-based windowing and pattern matrix similarity	Reduced false detection and improved accuracy
   20	Joaquim et al. (2022)	DBSCAN for smart greenhouse WSNs	Cluster-based anomaly detection using noise injection	Achieved 100% precision (2 features), 75.4% (5 features)
   21	Zhu et al. (2022)	Graph-based anomaly detection algorithm	Covariance matrix and high-correlation edge selection	Efficient anomaly detection with reduced computation
   22	Safaei et al. (2022)	Global outlier detection with time- series, entropy & RF	Time-series analysis + entropy + random forest	Detected anomalies with up to 99% accuracy
   23	Zakrzewski et al. (2022)	Logical sub-view based detection	Aggregator graphs for topology and traffic analysis	Detected anomalies from outside sub- views via aggregation
   24	Ifzarne et al. (2021)	ID-GOPA with online Passive Aggressive Classifier	Feature selection + online learning on WSN-DS dataset	Detected DoS attacks with up to 99% accuracy
   25	Mittal et al. (2021)	LMNN-integrated LEACH and IDS using SVM	Redesigned protocols and anomaly detection system	Achieved 96.15% accuracy in anomaly detection
   26	Biswas and Samanta (2021)	ERF for anomaly detection	Ensemble of DT, NB, and KNN; evaluated with AReM dataset	Outperformed individual classifiers
   27	Gavel et al. (2021)	Multilevel hybrid anomaly detection in IWSNs	SMO-SVM + OP-ELM with K-Medoid clustering	Achieved 94.79% accuracy, reduced false positives
   28	Arkan and Ahmadi (2021)	Entropy-based anomaly detection	Observation point relations + entropy + hierarchical topology	High accuracy in multilevel detection
   29	Lazar et al. (2021)	LSTM-based anomaly detection in SDWSNs	Autoencoder + LSTM trained on sensor data	Detected short- and long-term anomalies with 87% accuracy

3. MATHODOLOGY

   1. Threshold-Based Statistical Model

      Classification rule:

      ()

      () = +

      Let x(t) represent the sensor reading at time t. Define acceptable limits:

      ()

      An anomaly A(t) is detected if:

      = 1, (()) ( ) 0,

      Where w and b are determined by minimizing

      | | 2

      1

      | | +

      . .

      (

      + ) 1 ,

      () = 1, () < () >

      0,

      Where:

      2

      0

      • =

      • = +

      • =

      • =

      • : 2 3

   2. Moving Average Time-Series Model

      Given a window size w, the moving average is:

      1

      1

      1. Supervised Learning: Decision Tree

         The feature space is partitioned recursively. A sample is classified into a leaf node with majority class:

         () = 1, () =

         0,

         The decision tree optimizes splits by minimizing impurity:

         () = 1 2

         () =

         The absolute error is:

         ( )

         =0

         =1

      2. Unsupervised: PCA-Based Detection

         Anomaly if:

         () = () ()

         Data matrix ×, reduce to k-dimensional space:

         = where are top k eigenvectors Reconstruction:

         () = 1, () >

         0,

   3. Supervised Learning: Support Vector Machine (SVM) Given labeled data (, ) where {1, +1}:

   SVM finds hyperplane:

   Error: Anomaly if:

   =

   = 2

   ( ) = 1, >

   degree Celsius. Most of them fall within the typical operating

   0,

   range of 22°C to 23°C, indicating stable and normal environment.

   1. Unsupervised: Autoencoder-Based Detection

      An autoencoder is composed of encoder fand decoder g: Input: = [2, 1, ]

      Predicted value:

      +1 = ( ; )

      Error:

      e=1 +1

      Threshold derived from training set errors:

      = + 2

      Anomaly:

      ( + 1) = 1, >

      0,

4. TECHNIQUES FOR ANOMALY DETECTION

   1. Statistical Methods

      Statistical Techniques are the earliest and widely known techniques used for detection of anomalies in WSNs. These methods use p probabilistic models and statistical properties of the sensor data to determine deviations from normalcy. They are attractive for WSNs because of their simplicity and low computational cost.

      Threshold-Based Techniques

      The simplest method to identify anomalous behaviour is threshold-based anomaly detection by which thresholds are defined on sensor reading. If a value is greater than these thresholds, it is considered an anomaly. These thresholds may themselves be fixed (static) or adaptive (dynamically determined in response to observed patterns of recent data).

      Static thresholds are predetermined based on expert knowledge or historical data.

      Dynamic thresholds are more adaptive and can be calculated using moving averages, standard deviation, or percentiles.

      This technique is computationally efficient and easy to implement. However, it may struggle to detect complex or subtle anomalies, especially in non-stationary or noisy environments.

      Fig. 1. Threshold-Based Anomaly Detection

      The plotted figure depicts the result of thresholding-based anomaly detection on a set of temperature data gathered from a wireless sensor node. The x-axis is the order or number of sensors reading and the y-axis is the recorded temperature in

      The outlying two readings -45.0°C and 50.2°C are the extreme values when compared to the rest. These values are well above the previously set upper threshold of 30°C so it is considered as an abnormal behavior. Thus, we tag them as anomalies in Fig in form of red circles. These spikes might suggest a faulty sensor, interference or possibly a security attack generating false data. Lower (20°C) and upper (30°C) threshold limits are depicted as dashed red lines to guide in the identification of values that are considered out of the normal range. Temperature readings that are not within this range are identified as anomalies.

      It effectively shows that threshold-based detection is easy and quite efficient for clear outliers. But it also exposes one weakness: this approach would be indifferent to more subtle anomalies or irregular patterns that fall within the threshold range.

      Time Series Analysis

      Time series analysis considers the temporal dependencies and trends in sensor data. This method models the data as a sequence of values over time and detects anomalies by identifying points that deviate significantly from the predicted or expected pattern.

      • Moving average models (e.g., simple moving average, exponential moving average) smooth out short-term fluctuations and highlight longer-term trends.

      • Autoregressive models (e.g., ARIMA – AutoRegressive Integrated Moving Average) predict future values based on past observations. A large deviation between predicted and actual values may indicate an anomaly.

      • Seasonality detection can also be incorporated for periodic environmental conditions.

      Time series methods are more effective in capturing temporal anomalies, such as sudden spikes or gradual drifts in sensor readings, compared to threshold-based techniques. In a WSN deployed for environmental monitoring, if a humidity sensor shows a sudden drop not consistent with past hourly trends, the time series model may flag it as an anomaly, even if the value lies within general acceptable bounds.

      Fig. 2. Time Series-Based Anomaly Detection (Moving Average)

      The plot above is a moving average anomaly detection plot to capture temporal patterns in sensor data In the figure, fluctuations of sensor data are depicted by a new moving average method to detect an abnormal in a WSN. The Y-axis in above figure 2 represents the order of temperature readings over X axis as time, shows the actual temperature values captured by a sensor node. The black dashed line shows the estimated moving average, highlighting long-term trend and filtering out short-term variations. The solid blue line shows the actual observed values.

      That is, outliers are identified by calculating the difference between each sensor reading value and the moving average value for that sensor reading. If the deviation is greater than a predetermined threshold, the data point is characterized as Notorious. Red squares indicate these anomalies in the plot, i.e., there are certain points in which the sensor measurements differ a lot from the predicted trend according to historical ones.

      This technique is successful in detecting sudden and unexpected changes in sensor characteristics even though their values may be within the specified limits. For instance, if a sudden transition from 22.5°C temperature to 45.0°C is considered, it could be ignored by a static threshold (if set to 50°C) but would be anomaly according to the moving average model via its deviation from the recent average scheme. And a high reading, such as made at 50.2°C, which is in such marked contrast to preceding steady readings, is also called an anomaly. This figure illustrates the superiority of time seriesbased detection in detecting sudden changes, in addition to trends not according to normal patterns. But the interpretability of this approach critically hinges on the selected window size and

      deviation threshold.

      TABLE 2. ADVANTAGES AND LMITATIONS

      Advantages	Limitations
      Lightweight and resource- efficient	Poor performance in highly dynamic environments
      Easy to implement and interpret	May produce false positives due to noise
      Suitable for quick, online anomaly detection	Assumes statistical distribution and stationarity

      Statistical methods form a fundamental baseline for anomaly detection. However, their performance can be enhanced when combined with more advanced techniques such as machine learning and deep learning, which are discussed in the following sections.

   2. Machine Learning Approaches

      Supervised Learning Support Vector Machine (SVM) Supervised learning methods are extensively used in anomaly detection of WSN, in particular if labelled data are accessible. Among them, SVM is considered as one of the best classifiers based on its proficiency in taking account of high-dimensional features and drawing decision boundaries which can differentiate normal and anomalous data sets. SVM does this by searching for the hyperplane which best separates the data of two classes (usually normal versus anomaly) by a margin. In WSNs, we can utilize such SVM to learn from historical sensor data labelled as normal or abnormal and after the training it is able to classify new sensor data as normal and abnormal.

      Fig. 3. SVM based anomaly detection

      A Support Vector Machine (SVM) based anomaly detection technique is an example of supervised learning algorithm that aims to detect the outlier behaviors of the sensor nodes in WSNs. In this MATLAB simulation, simulated temperature data is applied to represent a typical WSN environment. We used 50 normal readings sampled at around 22°C, a few valid high temperature readings in the range of 35°C-50°C marked as anomalies in the training dataset. These labelled points are then used to train the SVM classifier to differentiate between normal and abnormal patterns.

      After training, the SVM model is used to test new sensor values, both for normal and anomalous excerpts. The classifier is correct in finding such anomalous readings e.g. 45°C,50°C to be outliers due to the decision boundary it has learned. The output is a binary classification for each test input, where the value '1' is the label for an anomaly and '0' for normal behavior. This classification can be visually confirmed as shown in the scatter plot, which distinctly separates the normal data points from the anomalies and highlights recently detected anomalies in red squares.

      Supervised Learning Decision Tree (DT)

      Decision Trees are intuitive, interpretable supervised learning classifiers. For WSN anomaly detection, they can be used to distinguish whether sensor readings are normal or abnormal according to the predetermined features, temperature, humidity, signal power, and so on. They recursively divide the data according to feature threshold to construct a tree-like model, so they are adapted for real-time decision and can be applied to low power networks such as WSN.

      Fig. 4. Decision Tree-Based Anomaly Detection

      The supervised-learning-based (Decision Tree) anomaly detection approach that classified the sensor data as normal or anomalous by following certain threshold (that had been learnt from labeled data) during the classification. By this implementation in MATLAB, synthetic node temperature that simulates readings for a WSN is used to the train the model. The training data contains 50 normal recordings with their mean at around 22 ° C and a few higher-temperature points according to figures presented i explicitly annotated as anomalies. This kind of variability helps in the decision tree to find patterns and differentiate between normal and abnormal behaviour.

      After being trained, the decision tree model is applied to new sensor readings, including normal and abnormal readings. The classifier correctly recognizes two reference temperatures 45°C and 50°C as abnormal, a proof of its ability to detect out- of- normal behavior, where the temperature values are much different from temperature values corresponding to regular operation.

      The main advantage of this strategy resides in the interpretability of the decision tree. Every decision node corresponds to a straightforward rule such as temperature > 30°C anomaly, which is easily interpretable and reliable to administrators or domain experts. In addition to that, decision trees are computationally very cheap and are optimal for real- time implementation in resource limited WSNs.

      The simulation using MATLAB will reveal that decision trees offer a pragmatic transparent and efficient approach to anomaly detection in WSN data specifically when dealing with structured numeric features; for instance, temperature. To improve accuracy for more complicated problems and achieve a more robustship, it is possible to extend the method to the case that more than one sensor information is availableor used as an ensemble method such as random forests.

      Clustering-Based Detection (K-means)

      Fig. 5. Clustering-Based Detection (K-means)

      PCA-Based Anomaly Detection

      Fig. 6. PCA-Based Anomaly Detection

      Autoencoder-Based Anomaly Detection (Shallow Autoencoder)

      Fig. 7. Autoencoder-Based Anomaly Detection (Shallow Autoencoder)

   3. Rule-Based and Heuristic Techniques

      In rule based/hueristic based method, explicit conditions or applications specific knowledge is used to identify the anamlies in a WSN. These methods are beneficial especially in low computational power systems with the need for lightweight processing and explicit logic. They are particularly appropriate for low power, resource-constrained WSN nodes for which complex ML or deep learning models are inapplicable.

      Application-Specific Detection Systems

      Beside these generic intrusion detection systems, application- specific anomaly detection systems are used that take behavior and expected patterns of the specific WSN deployment into account.

      Lightweight Detection in Low-Power Networks

      Wireless Sensor Networks consist in nodes with low battery and processing capabilities. In such environments, lightweight anomaly detection is very important. It is also about reducing computation: evaluating a few conditional statements is much more efficient than activating complicated models or transmitting data frequently.

      Common lightweight detection techniques include:

      • Static thresholding: Setting upper and lower bounds based on historical sensor data.

      • Sliding window analysis: Detecting anomalies based on deviation from the mean of recent readings.

      • Change detection rules: Monitoring the rate of change or variance in readings over time.

        These methods have minimal memory and CPU requirements, making them suitable for real-time on-node processing. Moreover, because they reduce the need for frequent communication with base stations, they also help extend the operational life of the network.

        Advantages

      • Requires no training or large datasets

      • Easy to deploy and update

      • Ideal for resource-constrained nodes Limitations

      • Limited generalization; highly dependent on domain knowledge

      • Cannot detect complex or subtle anomalies

      • May produce high false positives if thresholds are poorly defined

   4. Challenges in Anomaly Detection in WSNs

   In-network anomaly detection in wireless sensor networks (WSNs) is a challenging process that is affected by constraints of the network infrastructure (e.g. energy, and communication)and the variability of the network deployment environment. Although a variety of methods (including rule-based methods and deep learning models) have been proposed, different factors limit the practical utilization of the existing models.

   Resource Constraints (Energy, Memory, Bandwidth)

   Wireless sensor network (WSN) devices are usually small devices with limited computational capabilities and battery power supply. Running complex anomaly detection algorithms, especially machine learning or deep learning based, can consume node energy significantly and reduce the lifetime of the network.

   High False Positives

   There are many false positives in the A D system of WSNs, especially when data are noisy or changing. Light thin thresholding-based methods can cause the detection of large parts of normal variations as anomalous. High rate of false positives can overload the administrators, cause superfluous

   alarms, and degrade the trust of the detection system, hence decision-making may be affected and so are the responses.

   Dynamic and Heterogeneous Environments

   WSNs are often employed in varying environments such as forests, industrial areas, and farms in which sensor readings may be significantly different relative to environmental settings.

   Scalability and Real-Time Detection

   As WSN grow larger, to reach the size of hundreds or even thousands of nodes, anomaly detection systems need to evolve to that of large contents and decentralized deployments. On the other hand, centralized process becomes inefficient and could incur delays, and distributed detection needs to be light weight and synchronized among nodes.

   Data Imbalance and Noise

   The outliers of WSNs are naturally infrequent and result in imbalanced data. Models trained on similar data by machine learning not infrequently end up favouring normal behaviour too much and not being sensitive enough to actual anomalies.

5. CONCLUSION

   In this paper, the extensive review on anomaly detection methods in WSNs was provided that it is very significant for the sustainable, reliable, and efficient sensor-based monitoring systems security. Through comparing and analysing these works this experiment has shown that several techniques can be used and combined in combination in the constrains and requirement of the WSN. Results showed that there is possibility to use rule based and threshold based light weight methods for low power and resource limited environment, despite the fact that they have quite high false positive.

6. FUTURE WORK

In the future, it will be interesting to investigate adaptive, hybrid models which integrate the rule-based logic with learning-based methods to improve the accuracy of detection without adding much overhead to the computation requirement. The edge computing and federated learning can be combined to support decentralized learning while protecting data privacy and communication overhead.

REFERENCES

1. Revanesh, M., Gundal, S. S., Arunkumar, J. R., Josephson, P. J., Suhasini, S., & Devi, T. K. (2024). Artificial neural networks-based improved LevenbergMarquardt neural network for energy efficiency and anomaly detection in WSN. Wireless Networks, 30(6), 5613-5628.

2. CL, A., & Sumathi, R. (2024). Anomaly detection in WSN IoT (Internet of Things) environment through a consensus-based anomaly detection approach. Multimedia Tools and Applications, 83(20), 58915-58934.

3. Gayathri, S., & Surendran, D. (2024). Unified ensemble federated learning with cloud computing for online anomaly detection in energy-efficient wireless sensor networks. Journal of Cloud Computing, 13(1), 49.

4. Sharma, T., Balyan, A., & Singh, A. K. (2024). Machine Learning-Based Energy Optimization and Anomaly Detection for Heterogeneous Wireless Sensor Network. SN Computer Science, 5(6), 751.

5. Wei, Z. (2024). Simulation of Artificial Intelligence Algorithm Based on Network Anomaly Detection and Wireless Sensor Network in Sports Cardiopulmonary Monitoring System. Mobile Networks and Applications, 1-11.

6. Ahmad, R., & Alkhammash, E. H. (2024). Online Adaptive Kalman Filtering for Real-Time Anomaly Detection in Wireless Sensor Networks. Sensors, 24(15), 5046.

7. Xin, W., Ruixue, L., & Qing, L. (2024, December). Machine Learning- Powered Anomaly Detection in Wireless Sensor Networks: A Survey. In 2024 10th International Conference on Computer and Communications (ICCC) (pp. 1938-1944). IEEE.

8. Mohan, S., Manke, A., Verma, S., & Baskar, K. (2024). Machine Learning at the Edge: GANs for Anomaly Detection in Wireless Sensor Networks. In Enhancing Security in Public Spaces Through Generative Adversarial Networks (GANs) (pp. 305-317). IGI Global.

9. Haque, A., Chowdhury, N. U. R., Soliman, H., Hossen, M. S., Fatima, T., & Ahmed, I. (2023, September). Wireless sensor networks anomaly detection using machine learning: a survey. In Intelligent Systems Conference (pp. 491-506). Cham: Springer Nature Switzerland.

10. Chen, J., Zhang, J., Qian, R., Yuan, J., & Ren, Y. (2023). An anomaly detection method for wireless sensor networks based on the improved isolation forest. Applied Sciences, 13(2), 702.

11. Ravindra, C., Kounte, M. R., Lakshmaiah, G. S., & Prasad, V. N. (2023). Etelmad: anomaly detection using enhanced transient extreme machine learning system in wireless sensor networks. Wireless Personal Communications, 130(1), 21-41.

12. Kumar, A. S., Raja, S., Pritha, N., Raviraj, H., Lincy, R. B., & Rubia, J. J. (2023). An adaptive transformer model for anomaly detection in wireless sensor networks in real-time. Measurement: Sensors, 25, 100625.

13. Arul, J. S., & Venkatesan, R. (2023). A deep learning approach for efficient anomaly detection in WSNS. International Journal of Computers, Communications and Control, 18(1).

14. Srivastava, A., & Bharti, M. R. (2023). Hybrid machine learning model for anomaly detection in unlabelled data of wireless sensor networks. Wireless Personal Communications, 129(4), 2693-2710.

15. Premkumar, M., Ashokkumar, S. R., Jeevanantham, V., Mohanbabu, G., & AnuPallavi, S. (2023). Scalable and energy efficient cluster-based anomaly detection against denial of service attacks in wireless sensor networks. Wireless Personal Communications, 129(4), 2669-2691.

16. Mittal, M., Kobielnik, M., Gupta, S., Cheng, X., & Wozniak, M. (2022). An efficient quality of services based wireless sensor network for anomaly detection using soft computing approaches. Journal of Cloud Computing, 11(1), 70.

17. Yao, C., Yang, Y., Yin, K., & Yang, J. (2022). Traffic anomaly detection in wireless sensor networks based on principal component analysis and deep convolution neural network. IEEE Access, 10, 103136-103149.

18. Zhang, Q., Ye, M., & Deng, X. (2022). A novel anomaly detection method for multimodal WSN data flow via a dynamic graph neural network. Connection Science, 34(1), 1609-1637.

19. Wang, Z., Wang, Y., Gao, C., Wang, F., Lin, T., & Chen, Y. (2022). An adaptive sliding window for anomaly detection of time series in wireless sensor networks. Wireless Networks, 1-19.

20. Joaquim, M. M., Kamble, A. W., Misra, S., Badejo, J., & Agrawal, A. (2022). IoT and machine learning based anomaly detection in WSN for a smart greenhouse. In Data, Engineering and Applications: Select Proceedings of IDEA 2021 (pp. 421431). Singapore: Springer Nature Singapore.

21. Zhu, Q., Zhou, J., Zhao, S., & Wang, W. (2022, March). Graph-Based Anomaly Detection of Wireless Sensor Network. In Artificial Intelligence in China: Proceedings of the 3rd International Conference on Artificial Intelligence in China (pp. 144-150). Singapore: Springer Singapore.

22. Safaei, M., Driss, M., Boulila, W., Sundararajan, E. A., & Safaei, M. (2022). Global outliers detection in wireless sensor networks: A novel approach integrating timeseries analysis, entropy, and random forestbased classification. Software: Practice and Experience, 52(1), 277-295.

23. Zakrzewski, R., Martin, T., & Oikonomou, G. (2022, June). Anomaly detection in logical sub-views of WSNs. In 2022 IEEE Symposium on Computers and Communications (ISCC) (pp. 1-7). IEEE.

24. Ifzarne, S., Tabbaa, H., Hafidi, I., & Lamghari, N. (2021). Anomaly detection using machine learning techniques in wireless sensor networks. In Journal of Physics: Conference Series (Vol. 1743, No. 1, p. 012021). IOP Publishing.

25. Mittal, M., De Prado, R. P., Kawai, Y., Nakajima, S., & Muñoz-Expósito,

    J. E. (2021). Machine learning techniques for energy efficiency and anomaly detection in hybrid wireless sensor networks. Energies, 14(11), 3125.

26. Biswas, P., & Samanta, T. (2021). Anomaly detection using ensemble random forest in wireless sensor network. International Journal of Information Technology, 13(5), 2043-2052.

27. Gavel, S., Raghuvanshi, A. S., & Tiwari, S. (2021). A multilevel hybrid anomaly detection scheme for industrial wireless sensor networks. International Journal of Network Management, 31(4), e2144.

28. Arkan, A. S., & Ahmadi, M. (2021). Entropy-based anomaly detection using observation points relations in wireless sensor networks. Wireless Personal Communications, 119(2), 1783-1798.

29. Lazar, V., Buzura, S., Iancu, B., & Dadarlat, V. (2021, October). Anomaly detection in software defined wireless sensor networks using recurrent neural networks. In 2021 IEEE 17th International Conference on Intelligent Computer Communication and Processing (ICCP) (pp. 19-24). IEEE.