Leveraging Inter-Rater Agreement for Classification in the Presence of Noisy Labels