Аннотация
Today, the active implementation of machine learning (hereinafter – ML) methods in public administration opens up new opportunities for forecasting, impact assessment and decision support, while simultaneously generating various challenges. The present study is aimed at a systematic review of scientific publications devoted to applying ML methods in the field of public administration, with an emphasis on identifying thematic areas, ethical and institutional challenges. The initial data set included 524 publications obtained using targeted search queries in the Scopus and Web of Science databases for the period 2014-2024. Data filtering was performed using SQLite, thematic mapping was performed in the VOSviewer environment, and metadata was structured using the Elicit tool and subsequent manual encoding. The analysis results allowed us to identify four functional areas of ML application in public administration: transparency and ethics, resource allocation and service provision, institutional design, and technical integration. Despite significant progress in the models’ technical implementation and predictive accuracy, in many cases, mechanisms for equity, transparency, and citizen participation have been poorly implemented. The scientific novelty of the work lies in the interdisciplinary synthesis and development of a typology of institutional challenges that arise when implementing ML systems in public administration. The prospects for further research are related to the empirical validation of decisions, the development of ethical audit methods, and institutional training for responsible, sustainable, and contextually adaptive use of algorithmic tools in the public administration system.