Performance Evaluation and Comparison of Convolutional and Recurrent Neural Networks Using TensorFlow for Rip Current Hazard Prediction in Indonesian Tourist Beaches
CURRENT APPLIED SCIENCE AND TECHNOLOGY · 20.11.2026 · e0269383
DOI: 10.55003/cast.2026.269383 ↗Kurz gefasst
Fachartikel in der Zeitschrift CURRENT APPLIED SCIENCE AND TECHNOLOGY, online erschienen am 20.11.2026. Verfasst von Joko Sutopo und 4 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Rip currents are among the most dangerous coastal hazards, often linked to drowning incidents on tourist beaches worldwide. Early and accurate detection of rip currents is crucial for enhancing coastal safety and minimizing risks to visitors. This study evaluates and compares the performance of Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN), both implemented using TensorFlow, for predicting rip currents at three popular Indonesian beaches: Parangtritis, Drini, and Indrayanti. A dataset of 2,281 labeled drone images was prepared and divided into training, validation, and testing subsets for robust model evaluation. The CNN model used MobileNetV3-Large as the backbone to extract spatial patterns from coastal imagery. The RNN model incorporated a hybrid CNN–LSTM architecture designed to capture spatio-temporal dependencies, allowing it to detect sequential patterns that reflect the dynamic behavior of rip currents. Performance analysis showed that the CNN achieved consistent accuracy (0.88) across all sites, with precision ranging from 0.84 to 0.85, recall between 0.80 and 0.81, and AUC values up to 0.97. In contrast, the RNN achieved higher recall (0.83–0.87), indicating better sensitivity in detecting rip current regions, while maintaining AUC values between 0.89 and 0.92. However, its accuracy slightly decreased at Indrayanti Beach (0.845). These results show a trade-off: CNN excels in probabilistic discrimination, while RNN minimizes false negatives, making it more suitable for safety-critical early warning systems. This research contributes to AI-assisted monitoring systems, laying the groundwork for real-time detection frameworks to improve lifeguard decision-making and tourist beach safety.
Zitieren
APA
Sutopo, J., Suhirman, Wening, N., Aboobaider, B. M., & Rohmatika, F. A. (2026). Performance Evaluation and Comparison of Convolutional and Recurrent Neural Networks Using TensorFlow for Rip Current Hazard Prediction in Indonesian Tourist Beaches. CURRENT APPLIED SCIENCE AND TECHNOLOGY, e0269383. https://doi.org/10.55003/cast.2026.269383
Vancouver
Sutopo J, Suhirman, Wening N, Aboobaider BM, Rohmatika FA. Performance Evaluation and Comparison of Convolutional and Recurrent Neural Networks Using TensorFlow for Rip Current Hazard Prediction in Indonesian Tourist Beaches. CURRENT APPLIED SCIENCE AND TECHNOLOGY. 2026:e0269383. doi:10.55003/cast.2026.269383
BibTeX
@article{Sutopo2026,
author = {Sutopo, Joko and Suhirman and Wening, Nur and Aboobaider, ฺBurhanuddin Mohd and Rohmatika, Fiya Ainur},
title = {Performance Evaluation and Comparison of Convolutional and Recurrent Neural Networks Using TensorFlow for Rip Current Hazard Prediction in Indonesian Tourist Beaches},
journal = {CURRENT APPLIED SCIENCE AND TECHNOLOGY},
year = {2026},
pages = {e0269383},
doi = {10.55003/cast.2026.269383}
}
EUVIMED