Utilizing Long Short-Term Memory (LSTM) networks for predicting seismic-induced building damage: a Bawean Region case study

Zarkoni, Ahmad, Hariyadi, Mokhamad Amin ORCID: https://orcid.org/0000-0001-9327-7604, Almais, Agung Teguh Wibowo ORCID: https://orcid.org/0000-0001-7770-1954, Crysdian, Cahyo ORCID: https://orcid.org/0000-0002-7488-6217, Pagalay, Usman and Sugiharto, Tomy Ivan (2026) Utilizing Long Short-Term Memory (LSTM) networks for predicting seismic-induced building damage: a Bawean Region case study. Jurnal Ilmiah Teknologi Informasi Asia, 20 (1). pp. 8-15. ISSN 2580-8397

Preview

Text 28024.pdf - Published Version Available under License Creative Commons Attribution. Download (565kB) | Preview

Abstract

This study examines the feasibility of employing Long short-term memory (LSTM) networks to estimate earthquake-induced building damage using a focused dataset derived from the continuous 8-day mainshock–aftershock sequence that occurred in March 2024. A total of 483 events were analyzed, utilizing three readily available source parameters: magnitude, depth, and epicentral distance to predict the corresponding EMS-98 damage grade. The motivation for using an LSTM architecture stems from its capacity to model temporal dependencies in sequential seismic activity, despite the dataset's limited size. The best-performing single-split model (B4) achieved a test R^2 of 0.5738 and an RMSE of 0.2997 on the held-out set. However, to obtain a more robust assessment of the model’s generalizability, a 5-fold TimeSeriesSplit crossvalidation was conducted. The cross-validation procedure yielded a mean R^2 of 0.49 with a standard deviation of 0.27, and a mean RMSE of 0.33 with a standard deviation of 0.16. These results demonstrate that the LSTM model provides a credible baseline for exploratory damage estimation. However, a substantial portion of the variance remains unexplained due to the absence of geotechnical, soil-amplification, and structural-fragility information. The findings highlight the potential of sequence-based modeling for rapid damage estimation and underscore the need for integrating site-specific and structural variables in future work to enhance predictive accuracy.

Item Type:	Journal Article
Keywords:	disaster mitigation; lstm; prediction of building damage; seismic data; tectonic earthquakes
Subjects:	08 INFORMATION AND COMPUTING SCIENCES > 0801 Artificial Intelligence and Image Processing > 080105 Expert Systems 08 INFORMATION AND COMPUTING SCIENCES > 0801 Artificial Intelligence and Image Processing > 080199 Artificial Intelligence and Image Processing not elsewhere classified
Divisions:	Graduate Schools > Magister Programme > Graduate School of Informatics Engineering
Depositing User:	Mokhamad Amin Hariyadi
Date Deposited:	13 Jul 2026 09:22

Downloads

Origin of downloads

Actions (login required)

View Item