Comparison of the performance of different discriminant algorithms in analyte discrimination tasks using an array of carbon black-polymer composite vapor detectors

An array of 20 compositionally different carbon black-polymer composite che miresistor vapor detectors was challenged under laboratory conditions to di scriminate between a pair of extremely similar pure analytes (H2O and D2O), compositionally similar mixtures of pairs of compounds, and low concentrat ions of vapors of similar chemicals. Several discriminant algorithms were u tilized, including it nearest neighbors (kNN, with K = 1), linear discrimin ant analysis (LDA, or Fisher's linear discriminant), quadratic discriminant analysis (QDA), regularized discriminant analysis (RDA, a hybrid of LDA an d QDA), partial least squares, and soft independent modeling of class analo gy (SIMCA). H2O and D2O were perfectly classified by most of the discrimina nts when a separate training and test set was used. As expected, discrimina tion performance decreased as the analyte concentration decreased, and perf ormance decreased as the composition of the analyte mixtures became more si milar. RDA was the overall best-performing discriminant, and LDA was the be st-performing discriminant that did not require several cross-validations f or optimization.