@article{ART003370761},
author={Joon-Yong Kim},
title={Hybrid SARIMAX–LSTM Forecasting of Hourly Water Reduce Ratio with ROC-Based Risk Threshold Optimization},
journal={Journal of The Korea Society of Computer and Information},
issn={1598-849X},
year={2026},
volume={31},
number={8},
pages={147-157},
doi={10.9708/jksci.2026.31.08.147}
TY - JOUR
AU - Joon-Yong Kim
TI - Hybrid SARIMAX–LSTM Forecasting of Hourly Water Reduce Ratio with ROC-Based Risk Threshold Optimization
JO - Journal of The Korea Society of Computer and Information
PY - 2026
VL - 31
IS - 8
PB - The Korean Society Of Computer And Information
SP - 147
EP - 157
SN - 1598-849X
AB - Hourly water reduce ratio (WRR), defined here as the hourly rate of change in relative humidity, is used as an environmental proxy for rapid moisture-condition change relevant to fresh-concrete quality management. This study proposes a one-hour-ahead SARIMAX-LSTM hybrid forecasting framework in which SARIMAX models seasonal-linear dynamics and lagged weather effects, and LSTM corrects nonlinear residual components. The analysis uses 1,488 hourly observations from station ID 1189 between July 1 and September 1, 2021. Timestamp alignment, duplicate handling, missing-data treatment, and outlier auditing were explicitly specified. Wind speed was evaluated as a candidate exogenous variable, but it was not retained because it did not improve validation RMSE and MAE consistently. The final hybrid model achieved RMSE=4.413 and MAE=3.117 on the test set, representing numerical reductions of 1.66-5.99% in RMSE and 2.42-2.59% in MAE relative to the baselines. However, Diebold-Mariano tests with Holm correction and moving-block bootstrap confidence intervals did not establish statistical significance at the 0.05 level. ROC-based threshold optimization was conducted on the validation set using Youden's J and F1-score. Under WRR <= -5%, the selected alarm threshold was -0.479%/h and yielded Recall=0.926, Precision=0.210, F1=0.342, and ROC-AUC=0.788 on the test set. The framework is therefore suitable for early warning and secondary inspection support, not for direct diagnosis of concrete strength loss.
KW - Water Reduce Ratio;Time-Series Forecasting;SARIMAX;LSTM;Hybrid Model;ROC;Youden’s J;Threshold Optimization
DO - 10.9708/jksci.2026.31.08.147
ER -
Joon-Yong Kim. (2026). Hybrid SARIMAX–LSTM Forecasting of Hourly Water Reduce Ratio with ROC-Based Risk Threshold Optimization. Journal of The Korea Society of Computer and Information, 31(8), 147-157.
Joon-Yong Kim. 2026, "Hybrid SARIMAX–LSTM Forecasting of Hourly Water Reduce Ratio with ROC-Based Risk Threshold Optimization", Journal of The Korea Society of Computer and Information, vol.31, no.8 pp.147-157. Available from: doi:10.9708/jksci.2026.31.08.147
Joon-Yong Kim "Hybrid SARIMAX–LSTM Forecasting of Hourly Water Reduce Ratio with ROC-Based Risk Threshold Optimization" Journal of The Korea Society of Computer and Information 31.8 pp.147-157 (2026) : 147.
Joon-Yong Kim. Hybrid SARIMAX–LSTM Forecasting of Hourly Water Reduce Ratio with ROC-Based Risk Threshold Optimization. 2026; 31(8), 147-157. Available from: doi:10.9708/jksci.2026.31.08.147
Joon-Yong Kim. "Hybrid SARIMAX–LSTM Forecasting of Hourly Water Reduce Ratio with ROC-Based Risk Threshold Optimization" Journal of The Korea Society of Computer and Information 31, no.8 (2026) : 147-157.doi: 10.9708/jksci.2026.31.08.147
Joon-Yong Kim. Hybrid SARIMAX–LSTM Forecasting of Hourly Water Reduce Ratio with ROC-Based Risk Threshold Optimization. Journal of The Korea Society of Computer and Information, 31(8), 147-157. doi: 10.9708/jksci.2026.31.08.147
Joon-Yong Kim. Hybrid SARIMAX–LSTM Forecasting of Hourly Water Reduce Ratio with ROC-Based Risk Threshold Optimization. Journal of The Korea Society of Computer and Information. 2026; 31(8) 147-157. doi: 10.9708/jksci.2026.31.08.147
Joon-Yong Kim. Hybrid SARIMAX–LSTM Forecasting of Hourly Water Reduce Ratio with ROC-Based Risk Threshold Optimization. 2026; 31(8), 147-157. Available from: doi:10.9708/jksci.2026.31.08.147
Joon-Yong Kim. "Hybrid SARIMAX–LSTM Forecasting of Hourly Water Reduce Ratio with ROC-Based Risk Threshold Optimization" Journal of The Korea Society of Computer and Information 31, no.8 (2026) : 147-157.doi: 10.9708/jksci.2026.31.08.147
