[Context and motivation] Providing precise definitions of all project specific terms is a crucial task in requirements engineering. In order to support the glossary building process, many previous tools rely on the assumption that the requirements set has a certain level of quality. [Question/problem] Yet, the parallel detection and correction of quality weaknesses in the context of glossary terms is beneficial to requirements definition. In this paper, we focus on detection of uncontrolled usage of abbreviations by identification of abbreviation-expansion pair (AEP) candidates. [Principal ideas/results] We compare our feature-based approach (ILLOD) to other similarity measures to detect AEPs. It shows that feature-based methods are more accurate than syntactic and semantic similarity measures. The goal is to extend the glossary term extraction (GTE) and synonym clustering with AEP-specific methods. To evaluate our detection approach, the PROMISE requirements dataset is extended with 30 uncontrolled abbreviations without knowledge to the authors. ILLOD successfully extracted 28 of these abbreviations and correctly matched 25 pairs. [Contribution] In this paper, we present ILLOD, a novel feature-based approach to AEP detection and propose a workflow to its integration to clustering of glossary term candidates. First experiments show that ILLOD is well suited to augment previous term clusters with clusters that combine AEP candidates.