NEURAL-NETWORK AND EXPERIMENTAL-DESIGN TO INVESTIGATE THE EFFECT OF 5FACTORS IN ION-INTERACTION HIGH-PERFORMANCE LIQUID-CHROMATOGRAPHY

The effect of five experimental parameters on the ion-interaction chro matographic retention of pesticides characterized by different polarit y was investigated by means of experimental design and artificial neur al network treatments. The factors considered were: (1) the mobile pha se pH; (2) N, the alkyl-chain length of the IIR (ion-interaction reage nt); (3) CM, the organic modifier concentration in the mobile phase (4 ) CR, the concentration of IIR and (5) F, the flow-rate. The use of fr actional design and Hoke design allowed useful information to be drawn about the retention mechanism involved and to build, through artifici al neural network treatment (ANN), a model characterised by both descr iptive and predictive ability. Four neurons and a bias unit were emplo yed. The ANN proved to be a useful instrument in the optimisation of t he chromatographic separation, as regards resolution and total analysi s time: the experimental retention obtained in the optimal conditions always differed within 14% from the predicted ones. (C) 1998 Elsevier Science B.V.