Axions are well-motivated dark matter particles. Many experiments are looking for their experimental evidence. For haloscopes, the problem reduces to the identification of a peak above a noisy baseline. Its modeling, however, may be problematic. State-of-the-art analyses rely on the Savitzky-Golay (SG) filtering, which is intrinsically affected by any possible over fluctuation, leading to biased results. In this paper we study the efficiency that different extensions of SG can provide in the peak reconstruction in a standard haloscope-like experiment. We show that, once the correlations among bins are taken into account, there is no appreciable difference. The standard SG remains the advisable choice because of its numerical efficiency.