The learning machine in quantitative chemical analysis: Part 2. Potentiometric titrations of mixtures of three bases

Computer-calculated curves for the titration of mixtures of one strong base and two weak bases are used in the training and testing of a linear learning machine. The results indicate that multicategory classifiers can be calculated from a computer-generated training set of titration curves in which a random error of ±0.01 unit in the pH values is introduced. The relative error in the predictions for concentrations of bases not included in the training set was of the orderof ± 1% for concentration ratios up to 10:1 when δpK_b for the weak bases exceeded 1 pK unit and for K_b1 ? 5 × 10^-4 and K_b2 ? 10^-9. Calculation of the first derivative of the volume of titrant versus pH curves as a preprocessing step was necessary to obtain this accuracy for the weak bases, whereas the volume of titrant versus pH curves had to be used directly in the determination of the strong base. Predictions of concentrations of actual samples were in agreement with the computer-calculated results.