Comparative Analysis of Various Data Balancing Techniques for Propaganda Detection in Lithuanian News Articles

With the increased use of social networks, spreading propaganda has become more accessible, making developing propaganda detection methods crucial. However, only some datasets are available specifically for automatic propaganda detection, and the situation is particularly dire for low-resource languages such as Lithuanian. In this paper, balancing techniques are proposed as a solution to mitigate this limitation. Four balancing techniques, SMOTE, SMOTE-ENN, SMOTE-TOMEK and ADASYN, were evaluated for classification performance using LR, XGB, SVM, and RF models. The results showed that all data balancing techniques significantly improve the classification performance of machine learning models when the models are trained and tested on the same dataset. At the same time, SMOTE, SMOTE-TOMEK and ADASYN effectively improve performance when tested using new-unseen data, with the ADASYN technique proving to be superior.