Hybrid Multi-layer Perceptron and Metaheuristic Optimizers for Indoor Localization Error Estimation

Abstract

Indoor localization is hindered by GPS signal weakening in indoor environments. This research formulates machine learning with Multi-layer Perceptron Regression (MLPR) algorithm supported by two metaheuristic optimizers, namely, Gold Rush Optimizer (GRO) and Pelican Optimizer (POA), to yield hybrid models MLGR and MLPO to forecast Average Localization Error (ALE). The dataset organized in a structured form was of size 107 samples with six significant features as follows: anchor ratio, transmission range, node density, trainings, standard deviation of ALE, and ALE as objective. The dataset was split into training (70%), validation (15%), and testing (15%) subsets. Experimental analysis in three prediction layers reveals that MLGR outperformed MLPO and MLPR models in every prediction layer. MLGR exhibited maximum performance at the third test layer with an RMSE of 0.036 and R² of 0.993, whereas MLPO and MLPR attained RMSE of 0.059 and 0.080 and values of R² of 0.981 and 0.966, respectively. The findings establish the validity of the introduced hybrid optimization technique to increase accuracy and convergence rate of prediction of ALE in wireless sensor networks.