Formal concept analysis (FCA) is an important task for analyzing rela-tional data as well as extracting formalized knowledge from them. It has a wide range of practical applications, while a problem is frequently encountered that the extracted knowledge, i.e., the concept lattice may be too complicated for further processing and/or analysis. To reduce the concept lattice, various branches of techniques have been proposed and one of the most preferred categories of methods is the grouping-based simplification methods which aim at reducing the volume of the concept lattice by removing the redundant objects/attributes. However, we figure out that these methods may introduce too much information distortion to the context and the concept lattice, making the reduction result unreliable. To identify and overcome these problems, we propose a method for analyzing the reliability of a reduction in terms of context fidelity and lattice fidelity as well as a novel grouping-based concept lattice reduction method using an integer linear programming model. We conduct experiments on several data sets to prove that our method works in different cases and can produce more reliable reductions.