Modeling of Interpretable Fuzzy Rule-based System for the Classification of Thyroid Disease

Introduction: The ability of fuzzy rule-based systems to handle the imprecision and uncertainty present in real-world problems is quite impressive. The assessment of any fuzzy rule-based system in itself is a complex task because its performance is dependent on two parameters, namely accuracy and interpretability. Accuracy is assessed based on quantitative results, whereas interpretability is entirely subjective and may vary according to domain expertise. Methods: To develop an efficient fuzzy rule-based system, interpretability and accuracy should be taken into account, along with a good trade-off relationship between them. An interpretable fuzzy rule-based system has been developed in this paper for the classification of thyroid disease for an accurate diagnosis and prognosis of the thyroid. Result: The knowledge base for the classifier is built using three fuzzy rule induction algorithms (fuzzy decision tree, Wang Mendel, and fast prototyping) and two membership function generation schemes (k-means and HFP). Using these rule induction algorithms and membership functions, six different fuzzy-based classifiers have been suggested to deal with thyroid disorders, and their performance has been evaluated using accuracy and interpretability parameters. Conclusion: Among them, using hfp and kmeans partition with fuzzy decision tree rule induction algorithm, the classifier attained the highest accuracy of 95.8% and 93.5% with a good interpretability factor of 0.023 and 0.033, respectively, which shows a quite satisfactory result.