Evaluation of classification methods on imbalanced datasets

This thesis focuses on analyzing the performance of different classification methods when dealing with imbalanced datasets. It investigates the effect of data balancing techniques such as SMOTE, undersampling and oversampling on the performance of classification models. The main goal is to determine whether these techniques actually improve the prediction capabilities of classification models. The study focuses on three specific classification methods: logistic regression, Random Forest and Hellinger Distance Decision Tree (HDDT), and examines on which data they are most effective. It also analyzes the use of Leave-One-Out Cross-Validation (LOOCV) and thresholding using quantile as methods for evaluating models.