An Efficient Route Link Prediction Mechanism Using Link Appraisal Value (LAV) Based Manifold Perceptron Network (MPN) for Opportunistic Networks
Harshwin Venugopal, Jesu Jayarin Packiamani, Chandra Sekar Arumugam · Cybernetics & Systems · 2025
The proposed work focused on the learning rate optimization of Machine Perceptron Networks (MPNs) and improving efficient data communication in Opportunistic Networks (OppNets), which are characterized by their nature of being dynamically changing and unpredictable. The classic MPN optimization methods usually lead to poor performance for such networks when complex data classification problems are tackled. In addition, the link reliability estimations in OppNets are mostly suboptimal, resulting in inefficient communication, especially in environments where connectivity changes very frequently. These gaps need to be addressed in response in order to achieve better performance in machine learning tasks using MPNs and to stabilize the communication in OppNets. The two identified key gaps by this study are: there is an almost complete gap in effective optimization methods for enhancing the learning rate of MPNs to improve the accuracy of classifications; there are dynamic link reliability prediction deficiencies within OppNets that seriously limit efficiency in real applications. To solve these problems, this study puts forward the Elephant Herd Optimization (EHO) algorithm for MPNs’ learning rate dynamic adjustment to improve the accuracy of the model. Besides, a new link prediction mechanism based on LAV is proposed to estimate the link reliability according to the historical link behaviors and the Probability Density Function (PDF) of the previous links, combining with the Positive Probability of Link (PPL) defined as the probability of successful link establishment. This will further involve the optimization of the MPN learning rates using the EHO algorithm to achieve faster convergence and increase accuracy. Moreover, the link prediction mechanism based on LAV has about a 30% link failure rate decrease in simulation cases, which effectively shows the benefit of the method in enhancing the communication reliability for OppNets. The present results prove the proposed scheme, and point out that the combination of the EHO algorithm with LAV-based prediction would play a key role in significantly enhancing machine learning-based network communication under dynamic environments.