Abstract
In this study, machine learning methods were applied to a UCI dataset on colic in horses, consisting of 300 cases and 28 clinical variables. Preprocessing included cleaning, imputation, categorical variable encoding, and standardization, followed by class balancing through augmentation. Feature evaluation was performed using the inflation coefficient of variance to mitigate multicollinearity, recursive feature elimination to rank attributes, and principal component analysis to assess redundancy. Logistic regression, Random Forest, and XGBoost models were trained and evaluated. XGBoost achieved the best results with an accuracy of 87.0% and an F1 score of 87.1%, followed by Random Forest with 85.3% and logistic regression with 71.0%. The results demonstrate the potential of ensemble methods for predictive modeling of colic in horses.