ABSTRACT
In the current decade, lot of interest is expressed in healthcare applications using artificial intelligent (AI) systems. These systems indeed require considering explainable/interpretable machine learning algorithms that help physicians to analyze and validate extracted clinical information. To achieve interpretability as well as to deal with uncertainty, Fuzzy Decision Trees (FDT) are widely used in the decision-making process. The rules generated from decision trees quite interest the health care physicians to understand why a decision is made. This chapter mainly focuses on using FDT to express the benefits of explainable AI in health care domain. During the generation of FDT, each node is partitioned into specified number of clusters in general. Here, we propose to use cluster validity measure indices to determine the optimal number of partitions for each node of the tree. This alleviates the problem of choosing cluster numbers by a trial-and-error approach. In the experimental case study considered, HCV medical dataset from the UCI repository is used to construct FDT and to derive fuzzy rules. A total of eight FDTs are built with optimal fuzzy partitions for the nodes in the tree. Fuzzy c-means (FCM) clustering and cluster validity measurements are employed to select the optimum partitions that capture the intrinsic structure of the attribute values. Experiments on widely known medical data show that the proposed method can train compact and simple fuzzy rule bases and significantly provide good accuracy with an additional advantage of allowing physicians to interpret the decision-making process.
