The rise of social networks changed how people search for information and also facilitated fake news to spread; nowadays, fake news is often shorter, requiring specialized detection systems. In this paper, we proposed the Fuzzy XGBoost algorithm, which has three phases, to solve the multi-class fake news detection problem; this algorithm is a combination of Extreme Gradient Boosting models and a fuzzifier inspired by the Fuzzy c-Means Clustering algorithm. We experimented with the proposed algorithm on the LIAR dataset – a benchmark dataset for fake news detection, which has short statements and abundant meta-data; the experiment result proved that our proposed algorithm achieves higher accuracy compared to previous studies of fake news detection problems.