One major goal of data science applications is to extract patterns from large datasets. Such a database approach has been applied, for the first time in fusion science, towards a systematic study of the frequency spectra computed from reflectometry signals in the Tore Supra database, which are related to density fluctuations. In particular, the decomposition of reflectometry frequency spectra has allowed us to discover and quantify general trends of spectral characteristics (fractional contribution and width) for different components in Ohmic and low-confinement mode (L-mode) plasmas. In Ohmic plasmas, the contribution of the broadband component (EBB) to the total spectrum power in the saturated Ohmic confinement (SOC) regime is observed to be systematically higher than in the linear Ohmic confinement (LOC) regime. A transition of the dominating instability from TEM to ITG could explain the observed spectral modifications, which are supported by the analysis of the dependence of density peaking on collisionality. The spectral characteristics of the broadband (BB) and low-frequency (LF) components were then investigated in L-mode plasmas with ICRH or LH heating. The similar trends of the BB components with collisionality observed in L-mode plasmas compared with the Ohmic cases suggest a similar explanation by linking the frequency spectra to the underlying instabilities. The database analysis motivates more detailed studies by full-wave and gyrokinetic simulations, in order to confirm this link for both Ohmic and L-mode plasmas.