Dynamic data streams applications are bound to potential changes in data distribution, of which in the context of data stream mining, will cause concept drift. Data stream mining model must have the capability to adapt to concept drift, otherwise, risk the deterioration of its learning performance. Most of the existing surveys on concept drift detection methods focused on labeled data streams, that may be inapplicable to scenario where true labels are unavailable. The aim of this systematic literature review is to study the existing concept drift detection methods on unlabeled data streams, focusing on the learning process and the way concept drift is monitored in the data stream mining model. A total of 15 articles are selected for final analysis, and it is found that most of the drift detection methods were applied in a supervised learning setting. An experimental evaluation of the methods can be performed in the future work to investigate their performance in unsupervised learning setting. © 2020 IEEE.