Replicate-Level Experimental Data and R Code for Artificial Neural Network Modeling of Natural Rubber/Recovered Carbon Black Composites

Published: 2 November 2026| Version 1 | DOI: 10.17632/9b78yfhjn3.1
Contributors:
,
,
,
,
,
,
,
,
,
,

Description

This dataset provides comprehensive experimental measurements and computational scripts supporting the modeling of Standard Vietnamese Rubber 3L (SVR 3L) composites reinforced with hybrid blends of virgin carbon black (ASTM N220) and pyrolytic recovered carbon black (Polimix PX300). The total filler loading was kept constant at 50 phr while the vCB/rCB substitution ratio was systematically varied. To ensure computational reproducibility and meet peer-review requirements for rigorous model evaluation, this repository includes the complete replicate-level raw data and the R scripts used for the Artificial Neural Network (ANN). Crucially, to move beyond interpolation and demonstrate true predictive capability (generalization) without data leakage, this dataset also includes a genuine external validation partition, providing a robust reanalysis of the network's performance. Dataset Structure and Files: 1.RawData: 1.1.ANN_RawTrainingData.xlsx: Contains the 198 replicate-level experimental records (11 formulation levels across 18 replicates each) used strictly for internal training and ANN architecture selection. 1.2. ANN_RawValidationData.xlsx): Contains experimental replicate measurements of new, independent hybrid formulations synthesized exclusively for external validation of the model. These formulation ratios were entirely absent from the training set, serving as a strict held-out partition to rule out overfitting and prove model generalization. 2.ARED 2.1.AREDdata.xlsx: A comparative matrix containing the theoretical values predicted by the optimized ANN against the actual laboratory measurements from the external validation formulations. This file includes the step-by-step calculation of the Average Relative Error Deviation (ARED) for all rheometric and physicomechanical properties. 3.Codes 3.1.ANN_TrainingCode.R: The computational script written in R (using the neuralnet package) detailing the data partitioning, the random seed configuration (set.seed), and the resilient backpropagation training algorithms used to optimize the 1-[1,10,1]-9 multi-output architecture. 3.2.ANN_ValidationCode.R The selected ANN architecture that predicted physical-mechanical data for three distinct samples, which were subsequently validated through laboratory experiments.

Files

Institutions

Categories

Artificial Neural Network, Database

Licence