지방자치단체 수의계약의 연말 집중 현상과 점검대상 선별 가능성 연구 = Year-End Concentration of Non-competitive Contracts in Local Governments and the Feasibility of Selecting Audit Targets

다국어 초록 (Multilingual Abstract)

This study examines year-end concentration and heterogeneity in non-competitive contracting and proposes a data-driven screening approach using approximately 540,000 Korean local government contracts (2020-2025) from the Local Finance 365 Open API. Non-competitive status is coded from contract-method text, and periods are defined as year-start (Jan-Mar), year-end (Oct-Dec), and mid months. Year-controlled logistic regression indicates a significant year-end increase that recurs across years, with larger concentration for certain contract types and regions and strongest effects for contracts below KRW 100 million, especially below KRW 20 million. A screening model combining SBERT embeddings of contract titles and CatBoost shows high performance (ROC_AUC=0.977, PR_AUC=0.988, Brier=0.041). Cases predicted as competitive but observed as negotiated are treated as review candidates, reflecting both potential non-compliance and data/model limitations; linking additional information (e.g., failed bids, emergency grounds) is needed for validation.