<p>Seismology, tectonic history, and their importance in the region of Greece have been prominent throughout history due to the numerous recorded high-magnitude earthquakes that have resulted in severe damage and a significant number of casualties. In this work, a group of models created using a feed-forward neural network (FNN) is presented. These functions represent the power spectra of the spectral relative displacement (S_d), spectral relative velocity (S_v), and spectral maximum acceleration (S_a) of the response of a single-mass, single-damper, and single-spring system. Accelerometer recordings obtained in the region of Greece between 1975 and 2025 are collected, and the dynamic equilibrium equation is interpreted for varying values of the damping ratio (ξ) and eigenperiod (T) of the structure, to determine the corresponding spectral values. Following this, a matrix of the data of total size 30,300 was subdivided to obtain a spectrum that corresponds to an earthquake magnitude of 5.5, 6, 6.5, and 7 R. Subsequently, in each FNN model of a certain earthquake magnitude, the parameters of the input are ξ and T, while the parameters of the output are the spectral values. The models are constructed for the total of the components of the seismic acceleration direction: the direction corresponding to the East–West array, the direction corresponding to the North–South array, and the direction corresponding to the vertical Z array. The main results are as follows: the supervised learning rate of convergence is quick, requiring approximately 35 epochs, and the root mean square error is 0.03, which is approximately 1% in relative terms. The main eigenperiod interval detrimental to the structures in the study region is between 0 and 0.8&#xa0;s, corresponding to fairly stiff to middle-stiff buildings, which are the majority of the real structures in Greece. Subsequently, the probability of a resonance of a structure is increased in the study region.</p>