In this paper we present some soft computing methodologies for time-series
analysis applied to cyclostratigraphy. An application to some stratigraphic
signals to detect Earth orbital (Milankovic') periodicities which are expe
cted to be recorded in Cretaceous shallow water carbonate sequences outcrop
ping in Southern Apennines (Italy), is described. The results obtained with
classical spectral analysis techniques, based on the modified periodogram,
are compared to the results of our methods based on neural nets and geneti
c algorithms. The aim of these cross comparisons is to find the most reliab
le, fast and accurate methodology to identify orbital periodicities in nois
y and segmented stratigraphic signals. (C) 2001 Published by Elsevier Scien
ce Ltd.