FINDING AVALANCHE AREAS IN EAST KAZAKHSTAN USING MACHINE LEARNING

Abstract

Due to the high frequency and danger of avalanches in East Kazakhstan, the use of modern technologies for analyzing meteorological data has become critically important. This study focuses on the application of unsupervised machine learning algorithms - namely K-means and K-medoids—for detecting hidden patterns in weather conditions that precede avalanche events. The primary objective is to determine the most effective clustering approach for categorizing days with avalanche risk based on meteorological indicators such as temperature, snow depth, and weather conditions. The research involved comprehensive data preprocessing, conversion of categorical weather parameters into numerical values, and the selection of optimal cluster numbers using the Elbow and Silhouette methods. Experimental results indicate that the K-means algorithm outperforms K-medoids in clustering quality and interpretability. A total of four distinct clusters were identified, each reflecting unique weather patterns associated with avalanche activity. The results suggest that the applied methodology is suitable for future implementation in avalanche monitoring and early warning systems. This study contributes to disaster risk reduction efforts and supports decision-making processes for emergency response planning in mountainous regions.