Identification of Merapi Volcano Rockfall Direction Using Seismic Signals with Random Search Optimizer and K-Nearest Neighbors Method

Abstrak

This study aims to develop a system for identifying the direction of volcanic debris flows at Mount Merapi by applying the K-Nearest Neighbors (KNN) algorithm, enhanced with statistical feature extraction from seismic signals and optimized using the Random Search technique. The methodology involves three main stages: extracting statistical features from raw seismic data, implementing the KNN classification model, and tuning hyperparameters through Random Search to improve performance. The seismic dataset was sourced from monitoring stations operated by BPPTKG Yogyakarta. Model performance was assessed under varying training-testing data splits (70:30 and 80:20), K values, and cross-validation (CV) folds (3-fold and 5-fold). Results show that with a 70:30 split and K=1, the model attained an accuracy of 0.9054 across both CV types. When using an 80:20 split and K=2, accuracy improved to 0.9392, indicating that training data proportion significantly impacts model performance. However, variations in CV folds had negligible effects, suggesting that data ratio plays a more critical role in this context. This research contributes to the current state of the art by integrating the KNN algorithm with statistical feature extraction tailored for seismic signal analysis. Unlike previous studies, this work emphasizes parameter tuning via Random Search to enhance key performance metrics such as accuracy, precision, sensitivity, and AUC. The approach addresses limitations in previous models, particularly the challenge of determining the optimal K value, and provides a more robust method for forecasting volcanic flow directions using real seismic data from BPPTKG.

Identifier

• DOI: https://doi.org/10.62527/joiv.10.4.4536
• OpenAlex: https://openalex.org/W7202340695