Fuzzy decision trees in the support of breastfeeding

Abstract

Decision trees are a relatively well-known and often-used intelligent tool for decision support. They are convenient for capturing knowledge from vast amounts of data. The result (or 'decision model') is represented in a hierarchical manner, where the significance or contribution of a single attribute to the final decision is shown very clearly. When the decision tree is built on real-world data, this data is more often numeric than discrete. Because it is human nature to use words rather than numbers to describe something, fuzzy logic theory was introduced to fill this gap. The description of attribute properties using the fuzzy logic approach is represented by a plausibility vector, where the coordinates between 0 and 1 show the plausibility of an attribute belonging to one of its subsets of possible attribute values. This is the way to successfully overcome the problem of boundary values between attribute subsets, where sharply determined boundaries can greatly affect the final result. In our system design laboratory, we have developed a software tool for building decision trees with a fuzzy heuristic function. The tool was used on data collected from a breastfeeding booklet, and the results were used for developing different advisory systems for health-care professionals as well as for breastfeeding support groups and mothers who have access to the Internet.