A Novel Class Imbalance-oriented Polynomial Neural Network Algorithm for Disease Diagnosis

The class imbalance problem is common in disease diagnosis, which significantly damages the artificial intelligence-based diagnostic models and causes enormous cost about disease misclassification. To alleviate the class imbalance problem, we propose a novel class imbalance-oriented polynomial neural network (CIPNN) algorithm, which incorporates the data sampling and the classifier ensemble. Specifically, we utilize the nearest neighbor algorithm to identify a critical area from the given medical dataset (original dataset), which forms a critical area dataset. Then, we apply the synthetic minority over-sampling technique to generate a balanced dataset based on the original medical dataset. Further, we obtain the ensemble by combining multiple polynomial neural network classifiers, which are learned from the bootstrap samples sampled from the balanced dataset, the critical area dataset, and the original medical dataset, respectively. The experiments conducted on nine imbalanced medical datasets demonstrate that the proposed method can effectively alleviate the class imbalance problem.