Classification Approach towards Banking and Sorting Problems

Recently, ranking and sorting problems have attracted the attention of researchers in the machine learning community. By ranking, we refer to categorizing examples into one of K categories. On the other hand, sorting refers to coming up with the ordering of the data that agrees with some ground truth preference function. As against standard approaches of treating ranking as a multiclass classification problem, in this paper we argue that ranking/sorting problems can be solved by exploiting the inherent structure present in data. We present efficient formulations that enable the use of standard binary classification algorithms to solve these problems, however the structure is still captured in our formulations. We further show that our approach subsumes the various approaches that were developed in the past. We evaluate our algorithm on both synthetic datasets and for a real world image processing problem. The results obtained demonstrate the superiority of our algorithm over multiclass classification and other similar approaches for ranking/sorting data.