An outreach program is tasked with estimating the annual income of a household given its school district, neighborhood crime rate, proximity to clean drinking water, and roof type (there are three types: thatched, tin, or tile). The program only collected roof type information for a (uniformly random) 90% of households in its dataset and wants to impute (that is, ll in a best guess for) the missing roof type values prior to carrying out subsequent analyses. Suggest a reasonable way to impute the missing roof type values in a manner that makes use of the other features collected for each household.