We present an extension of selective editing techniques to detect potential influential errors for categorical variables using random forests. The proposal has been tested to rank the editing priority of the occupation variable in the Spanish branch of the European Health Survey under production conditions.