A procedure to detect suspected patterns of fraudulent behavior in vehicle emissions tests performed by an accredited inspection body

The National Institute of Metrology, Quality and Technology of Brazil (Inmetro), established by government decree 49/2010, has changed the requirements for vehicle safety inspections on both light and heavy vehicles that are converted to run on natural gas. In addition, according to government decree Inmetro 49/2010, the General Coordination for Accreditation of Inmetro is responsible for accrediting Brazilian vehicle safety inspection bodies. In recent years, there have been news reports, complaints, and denouncement about fraud cases in these accredited inspections, which increases the risk of accidents and environmental damage. In this paper, we propose a procedure to detect suspected fraud by an accredited vehicle safety inspection body. This tool combines clustering, digital analysis, and descriptive statistics to consider indicators of anomalous behavior and dataset object attributes in the clustering process. This mixed clustering structure links objects with similar anomaly scores together. We used descriptive statistics to identify which groups of observations were more likely to be fraudulent than others. In experiments, the proposed procedure identified unusual patterns successfully.