Fuzzy logika v predspracovaní údajov a jej vplyv na výkonnosť modelu strojového učenia XGBOOST

Abstract

Fuzzy logic provides an effective approach to preprocessing numerical data in machine learning, particularly in regression tasks. This paper explores the impact of fuzzification of variables such as age and BMI on the accuracy of healthcare cost prediction. By applying fuzzy transformation, we evaluated the performance of the XGBoost regressor across different dataset preprocessing variants. The results suggest that fuzzy logic can, in some cases, improve prediction accuracy (lower RMSE), especially for variables with unclear boundaries. We also discuss situations where its application does not lead to improvement and identify scenarios where it is most suitable.