Early Detection of Seismic Signal Anomalies Using Raspberry Pi 5 and Lightweight Machine Learning Models

Ahmad Kadarisman
Imam Fachruddin
Santoso Soekirno
Hanif Andi Nugraha
Benyamin Heryanto Rusanto
Martarizal

PROSIDING SEMINAR NASIONAL FISIKA (E-JOURNAL)
Vol. 14 No. 1 (2026): Joint Prosiding IPS dan Seminar Nasional Fisika

Data integrity is crucial for seismic monitoring systems, but is often compromised by anthropogenic or instrumental anomalies. This paper proposes a lightweight edge computing framework using Raspberry Pi 5 for real-time anomaly detection. MiniSEED data from the high-noise TOJI station were processed through segmentation, statistical or spectral feature extraction, and unsupervised models (isolation forest and autoencoder). The results show a detection latency of 78-113 ms with minimal resource consumption (<35% CPU, <200 MB RAM) and 82% correlation with ground-truth anomalies. This framework can be used on networked seismographs with limited resources such as those of the BMKG.