Akademik Veri Yönetim Sistemi
8th International Bursa Scientific Research Congress December 16-18, 2025, Bursa, TURKEY, 16 - 18 Aralık 2025, ss.852-860, (Tam Metin Bildiri)
ABSTRACT Leakage reactance is one of the key parameters influencing voltage regulation, short-circuit behavior and overall performance of single-phase transformers. Traditional analytical computation requires detailed geometric modeling and flux distribution assumptions, which may introduce uncertainties for practical coil structures. In this study, a Decision Tree regression model is developed to estimate the leakage reactance coefficient of simple cylindrical-coil monophase transformers using easily measurable geometric parameters. The dataset was generated synthetically based on realistic transformer design constraints and includes coil height, mean radius, insulation thickness, winding widths, and fictive iron height. The model was trained and evaluated in MATLAB Regression Learner using 5-fold cross validation. The Decision Tree achieved reasonably high predictive performance, with test R² of approximately 0.88 and low RMSE values. Shapley importance analysis revealed that coil height and mean radius had the dominant influence on leakage reactance estimation, consistent with electromagnetic theory. Predicted–actual comparisons showed a close alignment with the identity line, and residuals remained well distributed around zero. The results demonstrate that the Decision Tree approach provides a fast, interpretable, and engineering-oriented estimation tool suitable for preliminary transformer design stages. Keywords: Leakage Reactance; Monophase Transformer; Coil Geometry; Decision Tree; Machine Learning; Transformer Design.