Application of neural networks to empirical (vs. model) data: a diffusion tube experiment sample (white blood cells application)

The author describes the Mackey-Glass chaotic time series equations (1977) and the artificial neural networks that have been applied to (one of) them. The networks describe the time series model with sufficient accuracy. The question of whether the networks describe the corresponding physical phenomena with enough accuracy is then raised. A diffusion tube experiment to determine the gaseous diffusion coefficient of two components (CO/sub 2/ and air) and the experiment's corresponding analytical model are described. Artificial neural networks of 3-8-1 architecture are trained on empirical data from the diffusion tube system. The resulting network output, though extrapolated, is in the same accuracy range as the corresponding analytical model. A discussion of these results is included for the density of circulating mature white blood cells in a chronic granulocytic leukemia (CGL) patient.<
>