Detection of Fabricated Survey Data using Clustering Analysis

Data quality is the key of research successful. As more area of social sciences relay on the survey data, the quality of these data become more important. The data comes from the Questionnaires are used in the analysis of hypothesis that depend on the qualitative data. But, in a lots of situation, the answers are fraud according to different reasons. As a result, the analysis of these data leads to unrealistic hypotheses. Detection of this type of data can prevent the using of them and avoiding the incorrect results. Traditionally, statistics methods have been exploited to fraud data. But with machine learning evolving, more opportunity is available to provide solutions for problems with automated techniques. In this paper, clustering analysis would be suggested to detect the fake data based on learning process, clustering is central to many data-driven application domains. The goal of a clustering analysis is usually to obtain a symbolic description of the problem. Therefore, it would be used to describe the fraud data. The paper research work was based on training data, were given from two aspects. One dataset that is real data, were produced from real survey using more than 50 parameters. While another dataset is not real, it was produced fraudulently. Two clustering methods were used; k-means and k-mode according to their popularity and good results. The results uncover that fraud data has different behavior from real data. So it can be detected easily from this strange behavior.