Detection of Aberrant Responses in OMR Documents

摘要

Optical Mark Recognition (OMR) is a well-known process of capturing human-marked data in documents. It is an extremely accurate and rapid form of data capture especially when each response can be entered as a single mark and for the same reason this process has been employed in various examinations/tests across the world. It is clearly understood that conducting these tests in an equitable manner is of the utmost importance. Sadly, in the past few years, there have been several cases in which teachers/administrators of elementary and high schools across the United States were identified for fraudulently correcting the answers written by their students in order to improve the success rate of their respective schools. In order to identify this format of cheating, a procedure was developed to autonomously determine if cheating has occurred by detecting the presence of aberrant responses in scanned OMR test books. The challenges introduced by the significant imbalance in the numbers of typical and aberrant bubbles were identified. The aberrant bubble detection problem was formulated as an outlier detection problem. A pool of features is initially selected by examining bubbles that are penciled by a group of individuals and analyzing the differences between them. Several possible outlier detection methods were considered and a feature based procedure in conjunction with a one-class SVM classifier was developed. A multi-criteria rank-of-rank-sum technique was introduced to rank and select a subset of features from a pool of candidate features. Using the data set of 11 individuals, it was shown that a detection accuracy of over 90% is possible.