Application of machine learning method to analyse incomplete data

This paper presents an integrated approach to the analysis of incomplete and inaccurate data, illustrated by the example of mudflow forecasting. The aim of the study is to demonstrate how a combination of different methods allows not only to obtain adequate forecasts, but also to deeply understand the logic of decision-making by the model, identifying the key factors influencing the forecast. The key point of the work is the use of categorization of numerical data to increase the stability of models to outliers and noise, as well as to take into account nonlinear dependencies. The integrated approach is based on a combination of associative data analysis and the construction of a logical classifier, which acts as an interpreter of the obtained decisions. This combination made it possible to identify critical input features and understand how the model uses information to form a forecast, identify factors that have the greatest impact on the forecast result, ensure the accuracy and stability of forecasts taking into account the specificity and complexity of mudflow data. The rules obtained during the study, which are the key principles of the studied area, contribute to a deeper understanding of the nature of mudflows.