Predicting and recommending trust and distrust is undoubtedly crucial in social media. While these concepts may seem intuitive in real life, they are complex in online social networks. Hence the importance of understanding and predicting both trust and distrust in online social networks. However, while social studies have concluded that distrust is as important as trust, it is unfortunately ignored by the traditional approaches which consider it as the absence of trust. In this paper, we have enriched four real-world signed datasets with features such as reciprocity, notoriety, reputation, personal appreciation, and sociability. Using these preprocessed datasets, we put forth an implementation and conducted a comparative study of various prevalent supervised learning methods such as Polynomial regression (PR), K-Nearest Neighbors (KNN), Random Forest (RF), Category Boosting (CatBoost) for the prediction of trust and distrust values. Our comparative study shows that features implying a node’s own characteristics, and its direct neighbors, do provide simple, intuitive, and satisfying predictions.