Enhanced Distribution Matching for Multiclass Quantification

Quantification is the task of estimating the class distribution of a given dataset. This paper presents an Enhanced Distribution Matching method that can directly quantify multiclass datasets. Our method utilizes inter-class conditional probability distributions for quantification. We also propose a heuristic to speed up the distribution matching to obtain class distributions with a minimum number of iterations. The proposed methods were tested with 14 binary and multiclass datasets. The proposed method outperformed 12 datasets by a minimum 26% Mean Absolute Error (MAE) reduction.