RNN-Fuzzy-c-means-smote and fuzzy classifier to early predict hidden disease

Abstract

Unbalanced dataset classification issues have been prevalent in medical domain. Lately, clustering-based oversampling techniques were introduced to address this issue. Regrettably, they suffer from the vulnerability of hard grouping approaches. This paper introduces Recurrent Neural Network Fuzzy-C-Means SMOTE (RNNFCM-SMOTE) that balances data based on Recurrent Neural Network Fuzzy Fuzzy-C-Means in filtering phase. First of all, to manage the sensitiveness of the hard clustering, RNNFCM is used to determine safe regions. Second, take benefit of the capacity of neural networks to comprehend the features of data and dynamic systems to recover from past clusters, recurrent neural network is implemented to determine the membership function of different instances. To generate artificial data, classical smote is used. As the methods implementing fuzzy logic have proven to be very competent when it comes to the edge problem, RNNFCM-SMOTE is combined to nine fuzzy classifier methods to predict the existence of hidden diseases represented by five unbalanced medical data sets. The proposed method is compared to 12 oversampling methods using three performance measures. RNNFCM-SMOTE has been shown to consistently exceed many other popular oversampling techniques