Classic constitutive modeling of geomaterials based on the elasticity and p
lasticity theories suffers from limitations pertaining to formulation compl
exity idealization of behavior and excessive empirical parameters. This art
icle capitalizes on the modeling capabilities of neural networks as substit
utes for the classic approaches The neural network-based modeling overcomes
the difficulties encountered in understanding the underlying microscopic p
rocesses governing the material's behavior bf redirecting the efforts into
learning the cause-effect relations from behavioral examples. Several metho
dologies are presented and cross-compared for effectiveness in approximatin
g a theoretical hysteresis model resembling stress-strain behavior: The mos
t effective methodology was used in modeling the constitutive behavior of a
n experimentally tested soil and produced models that simulated the real be
havior of the soil with high accuracy. Although these models are empirical,
they are retrainable and thus, unlike classic constitutive modeling techni
ques, can be revised and generalized easily when new data become available.