Optimized Learning-Based Anomaly Recognition for Securing Wireless Sensor Networks

Abstract

Wireless sensor networks (WSNS) form the backbone of cybernez infrastructures by supporting critical applications such as environmental supervision, localization of objects and reliable data exchange. Their integration with the “Internet of medical things (IOMT)” and wider frames of the “Internet of Things (IoT)”, while offering increased functionality, also introduces significant vulnerability of cyber security due to an increased area of attack and heterogeneous ecosystem of the device. To solve these safety challenges, a framework for recognizing anomalies based on machine learning was developed to detect harmful behavior in WSN and IOMT environments. The framework emphasizes optimized learning through techniques of reducing advanced dimensions, using the analysis of the main components (PCA) and the decomposition of singular value (SVD) for selecting and extraction of elements. These techniques reduce the complexity of input data while maintaining critical information, improving the efficiency of the model and the accuracy of detection. The system's performance was verified using two benchmark data sets adapted to the attack scenarios in the contexts of WSN and IOMT. For classification, a hybrid voting classifier was implemented, integrated a stochastic gradient descent (SGD), bagging and strengthening decision -making trees to increase robustness and generalization. Access for learning uses the strengths of individual algorithms, while SGD offers scalability, reduces scattering through bootstrap and increases decision -making trees that improve performance through adaptive re -weighing incorrect cases. Experimental evaluation has shown the accuracy of 98.2% on the WSN data set and 73.8% on the IOMT data set, emphasizing the efficiency of the model when differential from legitimate activity in a heterogenic sensor environment. The methodology has shown that it is particularly suitable for remembered in real time by offering high accuracy, efficient calculation and resistance for dynamic attack formulas. This emphasizes the potential of sets of controlled, dimensionalized educational systems in ensuring modern wireless sensor infrastructures, especially when expanding to critical IoT applications, where data integrity and availability are paramount.