ASSESSMENT OF THE TRANSPORT AND OPERATIONAL CONDITION OF ROADS BASED ON MOBILE LABORATORY DATA USING MACHINE LEARNING METHODS

The subject of the study is the prediction of traffic intensity and pavement condition at a linear road section. The paper addresses a model of a neural network used to assess the usability of a road section and its transport and operational performance. The object of the study is a section of the M-1 Belarus road, 86th km, for the period from 2014 to 2023. The purpose of the study was to describe possible future scenarios of the road condition based on predicted traffic intensity and road quality condition metrics as part of the assessment of its usability with account for the International Roughness Index (IRI). In the course of the study, the following methods were used: Data Science (analysis of data collected from mobile laboratories) and machine learning algorithms (linear regression, gradient boosting, random forest, and neural networks based on long short-term memory (LSTM)). The output is a trained neural network capable of predicting the traffic intensity on the 86th km of the M-1 Belarus road. These methods reveal hidden patterns in the data and provide high-accuracy predictions. Results: The implementation of the deep learning model using the assessment of the condition of a linear road section will make it possible to address the main issues of road maintenance — to optimize time and reduce expenditures when planning and introducing measures at the stage of operation of transport infrastructure facilities, to take into account possible risks of road condition quality loss during re-pavement and design of new elements.