Decision making based on association rules

Data is being accumulated in a fast speed for many application domains, like finance and biology. Utilizing the huge volume of data to help make correct decisions is important for a company/organization to survive in this competitive world. Many general algorithms have been proposed in building decision-making systems. However, it is difficult to apply them to real-world domains without major changes due to different application natures (e.g. different goals, different data characteristics, etc). In this thesis, we study the problem of building decision-making systems using association rules for real-life applications. Unlike many existing algorithms that only touch the performance issue, we also focus on improving the interpretability of systems, which is very important in helping users understand how decisions are made (by the system). Association rules are easy to interpret and, thus, help us achieve this purpose. There are two major contributions in this thesis. First, we propose a common framework which can serve as the guideline for building decision-making systems. The design goal of this framework is to build both understandable and effective systems. To help make the system understandable, the framework uses association rules as the basic elements. Also it provides the flexibility for the users to prune the system using the domain knowledge. Such pruning is very important to keep the system small and, thus, understandable. To help make the system effective, it emphasizes pushing the application goal down to the rule searching phase (the first step of system building). As our second contribution, we propose a collection of algorithms for several real-life applications by following the guidelines in the framework. All proposed algorithms share the themes in framework; however, each of them is unique and is specially designed to meet the distinct challenges of its application domain. Experiments show the effectiveness of these algorithms.