Analysis of mudflow characteristics with limited data using machine learning models

In paper, a combined method for analyzing incomplete and distorted information is proposed, demonstrated by the example of mudflow forecasting. The main purpose of the study is to demonstrate the ability not only to create accurate forecasts, but also to analyze the decision-making mechanisms of the model, identifying significant parameters that affect predictions. To represent the identified sets of parameters affecting the volume of the mudflow in the form of logical rules, it was necessary to use data categorization. This made it possible to increase the reliability of models in the presence of emissions and noise, as well as to take into account non-linearities. Two approaches were used to form logical rules: the method of associative analysis and the original method of constructing a logical classifier. As a result of associative analysis, rules were identified that reflect certain patterns in the data, which, as it turned out, required significant correction. The use of a logical classifier made it possible to clarify and correct the patterns, ensuring the determination of a set of factors influencing the volume of mudflow. This approach made it possible to identify the most significant input variables and understand how the model processes data to generate a forecast, identify factors that play a key role in forecasting results, and ensure adequate accuracy and stability of forecasts, taking into account the specifics and complexity of mudflow data. The patterns deduced as a result of the study, reflecting the hidden principles of the subject area under study, and the methods of logical analysis used in the study helped to identify possible causes of the formation of different volumes of carried-out solid deposits. The results obtained can be used to improve monitoring systems and prevent the negative consequences of mudslides.

1. Caiafa C.F., et al. Decomposition Methods for Machine Learning with Small, Incomplete or Noisy Datasets. Applied Sciences. 2020;10(23). https://doi.org/10.3390/app10238481

2. Kainthura P., Sharma N. Hybrid machine learning approach for landslide prediction, Uttarakhand, India. Scientific Reports. 2022;12(1). https://doi.org/10.1038/s41598-022-22814-9

3. Hadi F.A.A., et al. Machine learning techniques for flood forecasting. Journal of Hydroinformatics. 2024;26(4):779–799. https://doi.org/10.2166/hydro.2024.208

4. Lombardo L., Mai P.M. Presenting logistic regression-based landslide susceptibility results. Engineering Geology. 2018;244:14–24. https://doi.org/10.1016/j.enggeo.2018.07.019

5. Rahmati O., Kornejady A., Samadi M., et al. PMT: New analytical framework for automated evaluation of geo-environmental modelling approaches. Science of The Total Environment. 2019;664:296–311. https://doi.org/10.1016/j.scitotenv.2019.02.017

6. Kondrat'eva N.V., Adzhiev A.Kh., Bekkiev M.Yu., Geduev (Gyaurgieva) M.M., Perov V.F., Razumov V.V., Seinova I.B., Khuchunaeva L.V. Kadastr selevoi opasnosti yuga evropeiskoi chasti Rossii. Moscow; Nalchik: Feoriya; 2015. 148 p. (In Russ.).

7. Kyul E.V., Ezaov A.K., Kankulova A.K. The theoretical basis of geo-environmental monitoring of the mountain geosystems. Sustainable Development of Mountain Territories. 2019;11(1):36–43. (In Russ.). https://doi.org/10.21177/1998-4502-2019-11-1-36-43

8. Radeev N.A. Avalanches Forecasting Using Machine Learning Methods. Vestnik NSU. Series: Information Technologies. 2021;19(2):92–101. (In Russ.). https://doi.org/10.25205/1818-7900-2021-19-2-92-101

9. Haoxiang Wang S.S. Big Data Analysis and Perturbation using Data Mining Algorithm. Journal of Soft Computing Paradigm. 2021;3(1):19–28.

10. Lyutikova L.A. Construction of a Logical-Algebraic Corrector to Increase the Adaptive Properties of the ΣΠ-Neuron. Journal of Mathematical Sciences. 2021;253(4):539–546. https://doi.org/10.1007/s10958-021-05251-3