The magnetic and velocity fields in sunspots are highly structured on small spatial scales which are encoded in the Stokes profiles. The Stokes profiles are in turn, derived from a sequence of polarization modulations on the incoming light that are imaged using an analyser-detector combination. Our aim is to identify Stokes profiles in a sunspot which exhibit spectral characteristics that deviate from those associated with the Evershed flow and their corresponding spatial distribution. To that end, we employ the k-means clustering routine to classify Stokes V spectra in the penumbra of a regular, unipolar sunspot, that also comprises a granular and a filamentary light bridge. We find that 75% of the penumbral region, corresponding to about 93500 pixels, is dominated by profiles comprising two, nearly anti-symmetric lobes, while 21% of the area is occupied by three-lobed profiles that are associated with the Evershed flow returning to the solar photosphere. The remaining 4% of the penumbral area is dominated by four groups/families of profiles - Group 1: three-lobed profiles in which both the rest and strong downflowing (sometimes supersonic) component have the same polarity as the host sunspot and seen exclusively in the filamentary light bridge. Group 2: single, red-lobed profiles occupying an area of about 2% located at the outer penumbra in discrete patches that possibly signify the downflowing leg of an Ω-loop. Group 3: three-lobed or highly asymmetric profiles, in which the rest component and the strong downflowing component have an opposite polarity as the sunspot. These occupy about 1.4% of the penumbra’s area and are seen in conspicuous, elongated structures or isolated patches in the outer penumbra and at the penumbra-quiet Sun boundary. Group 4: three lobed-profiles, in which the rest component has the same polarity as the sunspot and a weaker, upflowing component with a polarity opposite that of the sunspot. These profiles are located close to the entrance of the filamentary light bridge and are found in only 0.12% of the penumbral area. These minority groups of profiles could be related to dynamic phenomena that could also affect the overlying chromosphere. The simplicity and speed of k-means can be utilized to identify such anomalous profiles in larger data sets to ascertain their temporal evolution and the physical processes responsible for these inhomogeneities.