Kim Yee Lee0000-0002-1435-4163Yong-Hong Lee0000-0002-7551-0342Voon Hee WongEe Meng ChengGobi VetharatnamKok Yeow You0000-0001-5301-1115Eng-Hock Lim0000-0002-9365-0672Siong Kang Lim2026-10-222026-10-222025-11-1510.1007/s10921-025-01294-7https://dspace-cris.utar.edu.my/handle/123456789/11827This study presents a comparative evaluation of statistical and machine learning (ML) models for predicting moisture content (MC) in lightweight foamed concrete (LFC) using nondestructive microwave reflection parameters. Five models were examined: Multiple Linear Regression (MLR), Support Vector Machines (SVM), Random Forest (RF), Levenberg-Marquardt Neural Network (LMNN), and Radial Basis Function Network (RBF). Two dataset formats were used: a frequency-structured dataset (FSD) capturing full-spectrum information, and an instance-based dataset (IBD) designed for single-frequency applications. Model performance was assessed using R2, RMSE, MAE, training time, and prediction speed, with five-fold cross-validation applied to evaluate generalization. Results showed that ML models outperformed the statistical model in capturing non-linear relationships. RF and LMNN achieved the highest accuracy and stability across both datasets, while RBF and MLR showed signs of overfitting, especially on FSD. Sensitivity analysis using permutation feature importance revealed that S11 magnitude was most influential in structured data, whereas input importance was more evenly distributed in IBD. The findings emphasize the importance of aligning model choice with dataset structure to improve accuracy and robustness. This study supports the development of real-time, nondestructive moisture monitoring systems in LFC for sustainable construction applications. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2025.en-USDataset structureLightweight foamed concreteMachine learning modelMicrowave measurementsMoisture content predictionConcretesForecastingFunction evaluationImage retrievalLearning systemsLinear regressionMoistureNondestructive examinationRandom forestsSensitivity analysisSupport vector regressionFoamed concreteLevenberg-MarquardtMachine learning modelsMoisture content predictionsMultiple linear regressionsNon destructiveStatistical learningMicrowave measurementMoisture determinationRadial basis function networkstext::journal::journal article