GIS attribute data knowledge discovery system

This paper designs a system that discovers knowledge from a GIS attribute data list. The system can acquire useful knowledge from the attribute data in a spatial database and get if-then rules to assist in decision-making. Four models, a qualitative model, an importance-judging model, a decision-table reducing model and a decision-making model form the backbone of the system. Every model connects orderly. User inputs parameters in qualitative model by human-computer interaction. According to these parameters the system determines actual bell membership functions based on the golden section model. With the membership functions quantitative data are changed into qualitative data that form an initial decision-table. According to rough set theory we develop an importance-judging model and a decision-table reducing model. In the importance-judging model, based on indiscernible relation we get importance of every condition attribute to decision attribute. In the reducing model we provide an effective reducing method to get the most concise if-then rules. The reducing process consists of two steps: firstly forming a discernable matrix to get core attributes, then simplifying every rule until every rule is composed of the least condition attributes. In the decision-making model neural network is used to simulate the most concise rules getting from decision-table reducing model and test the general ability. At the same time the paper makes transparent three factors that affect the general ability. The paper presents an example of its use for judging drought and flood disasters in Songhua river basin. Simulation results show that the system can quickly form the most concise if-then rules and make the right decision.