Fiber to the home cluster development: Discovering a regression model to forecast fiber to the home take-up rate using fitness, bias/error performance, stability, and computational execution as model criteria
Year: 2026
Type: article
Abstract: The Take-Up Rate of customers in an area must be considered to ensure internet service provider performance when implementing public access via network or internet infrastructure. Previous studies have proposed several approaches for predicting Take-Up Rate (TUR) in FTTH deployments, including support vector regression (SVR), multiple regression, and multi-period network trajectory modeling. However, these methods often rely on limited time-series data and fail to establish clear relationships between input features and prediction outcomes. This research attempts to fill these gaps by creating and analyzing FTTH Cluster TUR prediction models utilizing a large dataset gathered in Indonesia between 2021 and 2024 that was influenced by eight internal and eleven external factors. We compare the performance of six models—Radial Basis Function (RBF), Extreme Learning Machine (ELM), ELM-RBF, ELM-RBF with Richardson Extrapolation, Artificial Neural Network (ANN), and SVR—on a dataset of 54,822 entries divided into three volume scenarios: full, quarter, and eighth, with training and testing ratios ranging from 70:30 to 80:20. Model Fitness, Model Bias, and Model Stability criteria use several evaluation measures, including R², MSE, RMSE, MAE, RRMSE, SMAPE, ESE, Friedman Test, and Nemenyi Test. Among all models, the ANN with Exponential Linear Unit (ELU) activation performed the best in terms of Model Fitness, Model Bias, Model Stability, and Computational Execution, as reflected by each evaluation. These findings demonstrate the predictive superiority of ANN for Take-Up Rate prediction modeling and then interpret the results to provide strategic benchmarks for cluster planning and homepass rollout optimization. While suggesting that ELM-RBF and SVR require additional tuning to obtain comparable accuracy.
Institution: Binus University
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7160703447