Fuzzy Clustering SMOTE and Fuzzy Classifiers for Hidden Disease Predictions

Medical data are generally imbalanced and have stochastic characteristics. Deterministic machine learning methods cannot correctly in such imprecise, uncertain, or fuzzy environments. Fuzzy reasoning techniques have shown to be more effective in learning from imbalanced or stochastic healthcare data. In this chapter, we focus on building automatic systems to predict hidden diseases based on Fuzzy C-Means SMOTE and fuzzy classifiers. The developed systems were tested on several medical datasets and were compared against hard prediction approaches that implement hard oversampling methods and classifiers. With respect to the performance measures geometric mean, harmonic mean, and area under curve, [Fuzzy C-Means SMOTE + Fuzzy KNN] and [fuzzy C-Means SMOTE + Fuzzy Decision Tree] outperformed the deterministic systems and showed a great ability to generalize their learning to unseen patients.