The attribute reduction of a three-way concept lattice preserves the structure of the entire lattice. The reduct can be used to simplify the three-way formal concepts as a whole from the global lattice perspective. As a result, attribute reduction becomes one of the core topics in three-way formal concept analysis, which contributes to the learning, interpretation, and application of knowledge discovered from a formal context. Preserving the whole lattice structure may be unnecessarily strong in certain applications such as concept-cognitive learning. In those cases, it may be sufficient and meaningful to preserve certain learning properties with respect to an individual concept. Therefore, this work proposes and explores attribute reduction of a single three-way formal concept, which preserves the existence of either an object-induced or attribute-induced concept with the same extension or intension respectively. Firstly, we propose the definition of a concept-oriented reduct and the approaches to constructing such a reduct. Secondly, we explore the connections between the proposed concept-oriented reducts and the existing lattice-oriented reduct. Thirdly, we discuss the connections and differences between the attribute trisections produced from the two types of reducts. Lastly, two algorithms are introduced to construct a concept-oriented reduct, whose effectiveness and efficiency are illustrated by experiments with UCI datasets.