[kən′fyü·zhən ‚mā·triks] (computer science) In pattern recognition, a matrix used to represent errors in assigning classes to observed patterns in which the ij th element represents the number of samples from class i which were classified as class j.