In recent years, complex machine learning models have been widely introduced in various industrial fields due to their high accuracy. However, their increasing complexity has been a major obstacle to their implementation in sensitive decision-making situations. In order to address this problem, various post-hoc explanation methods have been proposed, but they have not been able to achieve interpretability of both the explanation and its scope. We propose R-LIME, a novel method that interprets complex classifiers within an interpretable scope. R-LIME locally and linearly approximates the complex decision boundary of a black-box classifier within a rectangular region and maximizes the region as long as the approximation accuracy exceeds a given threshold. The resulting rectangular region is interpretable for users because it is expressed as a conjunction of feature predicates. Through qualitative and quantitative comparisons on a real-world dataset, we demonstrate that R-LIME provides more reliable and interpretable explanations than existing methods.