COMPARISON OF PATTERN-RECOGNITION TECHNIQUES FOR SAMPLE CLASSIFICATION USING ELEMENTAL COMPOSITION - APPLICATIONS FOR ICP-AES

Pattern recognition is very important for many aspects of data analysi s and robotic control. Three pattern recognition techniques were exami ned-k-Nearest Neighbors, Bayesian analysis, and the C45 inductive lear ning algorithm. Their abilities to classify 71 different reference mat erials were compared. Each training and test example consisted of 79 d ifferent elemental concentrations. Different data sets were generated with relative standard deviations of 1, 3, 5, 10, 30, 100 and 500%. Ea ch data set consisted of 2000 examples. These sets were used in both t he training stages and in the test stages. It was found that C4.5's in ductive learning algorithm had ii higher classification accuracy than either Bayesian or R-Nearest Neighbors techniques, especially when lar ge amounts of noise were present in the systems.