Optimizing MANET Performance: A Machine Learning Solution for Achieving 92%+ Accurate Signal-To-Noise Ratio Predictions in Dynamic Environments

The MANETs are applied in dynamic environments where accurate estimation of Signalto-Noise Ratio (SNR) is required for communication.Traditional statistical and measurement-based SNR estimation methods prove inadequate for real-time adaptation to rapidly changing network conditions.This paper introduces a novel machine learning regression framework using Gradient Boosting Regressor (GBR) for enhanced SNR prediction in MANETs, validated through comprehensive field measurements and robust statistical analysis.A hybrid dataset of 600 samples was constructed, comprising 75% real-world field measurements from urban, rural, and emergency response scenarios, and 25% validated simulations.The optimized GBR model, configured with 150 estimators, 0.1 learning rate, and maximum depth of 5, was evaluated using repeated 105-fold cross-validation.Results demonstrate superior performance with R = 0.914 0.018, MAE = 0.58 0.09 dB, and MSE = 0.49 0.12 dB, significantly outperforming Random Forest (R = 0.847 0.024) and Linear Regression (R = 0.743 0.031).Comprehensive validation including learning curve analysis, bootstrapping with 1000 samples, and field measurement correlation (r = 0.887) confirms model generalizability and absence of overfitting.The framework demonstrates robustness across diverse mobility scenarios and provides a scalable, real-time solution for intelligent MANET optimization in emergency response, vehicular networks, and military communications.