Predicting the characteristics of sele flows using machine learning

Mudflows are among the most hazardous natural phenomena in mountainous regions, causing significant damage to infrastructure and settlements. Their prediction is traditionally challenging due to the fragmented, incomplete, and heterogeneous nature of source data, which limits the applicability of classical statistical methods. Purpose of the research. To develop and test a set of machine learning models for predicting mudflow characteristics and identifying the factors determining their intensity, based on incomplete data. Materials and Methods. Using the mudflow cadastre of the North Caucasus, descriptive, correlation, and cluster analyses were applied. To overcome data nonlinearity, discretization of numerical features was performed, followed by a transition to classification tasks and associative analysis using CART and FP-Growth algorithms. Results. Clustering revealed three physically interpretable groups of mudflows. The classification model based on discretized data demonstrated 100 % accuracy in predicting debris flow volume. Association analysis identified robust logical rules linking flow genesis, basin morphometry, and solid deposit volume. Discussion. It was established that the largest discharge volumes are characteristic of glacial-rainfall-induced mudflows in large basins. Data discretization enabled a shift to interpretable models that outperformed traditional linear methods in accuracy. Conclusion. The effectiveness of machine learning methods for analyzing mudflow processes under conditions of incomplete data has been confirmed. Resume. The proposed approach enables the construction of highly accurate and interpretable predictive models. The results reveal key factors influencing the volume of mudflow discharge. The obtained rules can be implemented in mudflow monitoring and early warning systems. Future work should focus on enhancing the models by incorporating meteorological and lithological data

Cluster analysis · Data set · Debris flow · Discretization · Mudflow · Sorting · Climate change and permafrost · Karst Systems and Hydrogeology · Landslides and related hazards