| Abstrak/Abstract |
The southern area of Java Island has a significant
potential for tsunami events due to its position close to the ring
of fire and the megathrust tsunami potential of its seismic gap
study. A tsunami warning based on its category has become a
solid standard in Tsunami Community Preparedness for
Tsunami Disaster Prevention. The widely accepted method for
providing such Tsunami Warning Information is a non-linear
tsunami simulation that requires a vast computation resource.
In contrast, the tsunami travel time to the coast is short. Hence,
the available time to issue the warning is minimal.
This study aims to introduce the Artificial Neural Network
(ANN) to alternately classify the Tsunami Warning Level in the
Southern Area of Java Island with the case study using the
Hypothetical Earthquake of Java Megathrust (Mw 6.0-8.8). The
ANN applied to one of the models with the highest test accuracy
picked from the available configuration model, combining two
or six hidden layers, four activation functions, and two epochs.
The investigations showed that 6HL-Leaky-RELU-18E is the
chosen configuration model and had an excellent performance
with a validation accuracy of 100%; and Precision, Recall, F1
Score of 1.00, 0.79, 0.87, respectively. When tested with the
external dataset, it generated a test accuracy of 82%, with
91.67% of site/tsunami warning status correctly predicted. |
