Hybrid Levenberg–Marquardt and LSBoosting Ensemble Algorithms for Optimal Signal Attenuation Modeling and Coverage Analysis

Abstract

This paper proposes and engages the Levenberg–Marquardt algorithm method via regression to optimally model and predict real-time signal strength values acquired via telecom software investigation tools in LTE cellular networks. To further improve the Levenberg–Marquardt method, which is sometimes prone to parameter evaporation on high dimensional data with high bias or variance issues during the application, we explore the Least Square Boosting (LSBoosting) ensemble algorithms. The combined signal predictive modeling procedure is termed the hybrid LSBoost-LM method. First, when the proposed hybrid LSBoost-LM method was engaged for real-time extrapolative signal analysis, the results displayed excellent root mean square error precision accuracies compared with two other standards, Bag-LM and LM methods. As a case in point, the LSBoost-LM method achieved 2.15, 3.44, 3.33, 1.31, and 2.19 dB RMSE values at different prediction study locations, which are relatively lower compared with the Bag-LM and standard LM methods that achieved higher RMSE values of 3.38, 3.86, 4.07, 2.28, 3.98 dB and 5.57, 5.52, 5.14, 3.67, 4.56 dB, respectively. Secondly, applying the hybridized model produced up to 93.39% cell area coverage quality and 89.18% fringe cell area coverage quality across the eNodeB study locations. The proposed method can assist practicing RF network planners in realistic cell coverage quality estimation and analysis of related wireless networks.