Measurement errors in pattern recognition

The effect of measurement errors on the performance of the pattern classifier is studied. It is shown that measurement errors affect the optimality of the decision rule, and can result in higher classification error probability. The results derived in the paper are applied to the problem of quality control of gas meters. It is shown that, to maintain the same quality level of lots of 300 meters, the test sample set size had to be increased from 100 to 160 meters.