Towards Rough Set Theory for Outliers Detection in Questionnaire Data

Manual processing of questionnaire surveys takes a lot of time and effort. This article aims at the automatic detection of corrupted or inappropriate responses in questionnaire data using unsupervised outliers detection methods. Unlike numerical data, which are usually assessed by distance-based methods, the entries in questionnaires need to be assessed from multiple perspectives. This paper proposes a novel algorithm utilizing the rough sets that capture relations among attributes/questions. The rough set theory is based on the granularity of data and is used to find combinations of attributes identifying the discernible questionnaires. The method is compared with standard and recent outlier detection algorithms that are based on distance, entropy, correlation, and probability. The tests are computed on the real-world HBSC dataset using several experiments. The rough set score computed on combinations of three attributes is preferred as it returns significant outliers that even reflect multiple perspectives investigated by other types of methods.