Résumé : Radio used as a star formation rate (SFR) tracer presents enormous advantages by being unaffected by dust and radio sources being pinpointed at the sub-arc-second level. The interpretation of the low frequency 1.4 GHz luminosity is hampered by the difficulty in modeling the cosmic ray paths in the interstellar medium, and their interactions with the magnetic field. In this work, we compare the SFR derived from radio observations, and the ones derived from spectral energy distribution (SED) modeling. We aim at better understand the behavior of the SFR radio tracer, with a specific emphasis on the link with star-formation histories. We used the SED modeling code Code Investigating GALaxy Emission, CIGALE, with a non-parametric star formation history model (SFH) and fit the data over the wavelength range from the ultraviolet (UV) up to the mid-infrared (mid-IR). We interpret the difference between radio and SED-based SFR tracers in the light of recent gradients in the derived SFH. To validate the robustness of the results, we checked for any remaining active galaxy nuclei (AGN) contribution and tested the impact of our SFH modeling approach. Approximately 27% our galaxies present a radio SFR (SFR$_{\rm radio}$) at least ten times larger than the instantaneous SFR from SED-fitting (SFR$_{\rm SED}$). This trend affects primarily the galaxies that show a declining SFH activity over the last 300 Myr. Both SFR indicators converge toward a consistent value, when the SFHs are averaged over a period larger than 150 Myr to derive SFR$_{\rm SED}$. Although the radio at low frequency 1.4 GHz is a good tracer of the star formation activity of galaxies with constant or increasing SFH, our results indicate that this is not the case for galaxies that are quenching. Our analysis suggests that the star formation time sensitivity of the radio low frequency could be longer than 150 Myr.