USING MACHINE LEARNING METHODS TO MODEL ICE DATES ON RIVERS ON THE LAKE SEVAN BASIN

Abstract

The study proposes a solution to the problem of modifying ice forecasting, choosing two popular approaches: decision trees and a Naive Bayes classifier. The study focuses on rivers in the Lake Sevan basin, specifically the Dzknaget, Drakhtik, Pambak, Vardenis, and Bakhtak, which experience complex ice breakup and freezing processes. To schematize and combine the ice formation and breakup processes, dummy variables are used as forecast characteristics instead of the onset and end dates of ice events, with average ten-day air temperatures and water levels serving as independent parameters. According to satisfactory results, the proposed models perform less well in determining the end date of ice events; however, they are capable of handling small samples and the absence of functional relationships.