Methodology for Predicting Work on the Maintenance and Repair of Urban Facilities Using Machine Learning

This research work is devoted to the development of a methodology for predicting work on the maintenance and repair of urban facilities using machine learning. The initial data was collected from various sources of appeals for buildings in Moscow. Data processing was carried out, a dataset with new attribute fields was formed, an exploratory data analysis was carried out, where the main task was to identify data dependencies. The results of the study made it possible to form a hypothesis about the relationship that certain types of repairs can statistically reduce the number of incidents recorded by various sources. Hypothesis testing was carried out using cohort data analysis and A/B testing. Next was the stage of developing ML models to predict the level of importance of performing various types of repairs at a particular facility in terms of reducing the number of incidents at this facility and directly predicting repairs for facilities. After testing the hypothesis and validating the model on an example, a methodology was developed for predicting work on the maintenance and repair of urban facilities.