A Cost Sensitive and Class-Imbalance Classification Method Based on Neural Network for Disease Diagnosis

Abstract

The automation of disease diagnosis is confronted with three important problems which are class imbalance, sampling bias and cost sensitivity. In order to make a reasonable representation of the imbalance state, class distribution histogram and likelihood are devoted to measuring degree of its imbalance. A cost optimization model for disease diagnosis is proposed, which be successfully used in disease diagnosis and significantly reduce the negative effects of the above three factors.