Data augmentation-aided machine learning prediction of 28-day compressive strength of CNT/cement composites

Accurately predicting the 28-day compressive strength (CS) of carbon nanotube-reinforced cement composites (CNTRCCs) is crucial for accelerating their application in civil engineering. However, manual data collection for CNTRCCs is a time-consuming and labor-intensive process, further exacerbated by the relatively small quantity of available data. In light of these challenges, this work proposes an augmented technique called improved conditional tabular generative adversarial network (ICTGAN). The primary objective of ICTGAN is to augment the original training set, with the ultimate aim of improving the accuracy of the machine learning (ML) model in predicting CS of CNTRCCs. Furthermore, this study employs natural language processing (NLP) and multilayer perceptron (MLP) techniques to extract precise information from non-numeric features and integrate them into the ML models. The findings demonstrate that the most significant enhancement effect is observed when the ratio of the generated datasets to the original datasets is 0.5, with R² being 0.954. Simultaneously, employing NLP and MLP techniques to extract non-numeric feature information is found to be superior to traditional code methods. Additionally, the AutoGluon (AG) framework used in this work enables faster model deployment than traditional ML workflows and avoids the time-consuming process of tuning hyperparameters.