A Supervised Machine Learning Approach: Towards The Automatic Classification of Infrasound Events | ||
| Engineering Research Journal (Shoubra) | ||
| Articles in Press, Accepted Manuscript, Available Online from 23 March 2025 | ||
| Document Type: Research articles | ||
| DOI: 10.21608/erjsh.2025.353192.1392 | ||
| Authors | ||
| Eman Lotfy Habbak* 1; Mohamed Salah Abdalzaher2; Hala HA Mansour3 | ||
| 1ENDC Department, National Research Institute of Astronomy and Geophysics, Cairo 11421, Egypt. | ||
| 2Department of Seismology, National Research Institute of Astronomy and Geophysics, Cairo 11421, Egypt. | ||
| 3Department of Electrical Engineering, Faculty of Engineering at Shoubra, Benha University, Cairo, Egypt. | ||
| Abstract | ||
| Infrasound, which refers to low-frequency sound waves below 20 Hz, originates from various natural and human-made sources. These signals travel through a dynamic atmosphere that might change within minutes of an incident, so their classification process is challenging and sometimes time-consuming. Moreover, accurate classification of infrasound events is crucial for monitoring nuclear test bans and detecting natural disasters. Recently, the use of machine learning (ML) for complex environments and different signals has been merged. This paper presents a supervised ML approach to classify infrasound signals, utilizing feature selection methods - “SelectKBest” and “SelectFromModel”- and feature importance to identify the eight most relevant features of these signals. Traditional machine learning methods were selected over deep learning due to their interpretability, lower computational cost, and effectiveness in handling the available dataset size and variability. The model is trained and examined by a real dataset from the infrasound reference event database (IRED) in time and frequency domains, which are processed using the progressive multi-channel correlation (PMCC) algorithm. Throughout, the system model, feature selection is adopted to use only eight features to reduce the complexity. To ensure the models’ robustness, we have examined them by several evaluation metrics. The results show the model’s effectiveness in the events classification with an accuracy of 92.87% among the benchmarks ensuring the capability of ML for automatic classification. The proposed framework demonstrates significant potential for real-world applications, particularly in nuclear monitoring and natural disaster prediction, where timely and accurate decision-making is critical. | ||
| Keywords | ||
| Event discrimination; Real-time monitoring; Earthquakes; Explosions; Artificial Intelligence | ||
|
Statistics Article View: 35 |
||
