Dynamic Information System Decision Rules Mining Model And Application

In rules mining theory, the object traditionally processed is a static information system at a time, therefore the acquired knowledge is also static. This is different from actual information systems, which are always inconstant and procedural. In order to depict the dynamic change trend and rule of the information system, this paper proposes the dynamic information system transformation model and provides its representation formalism, through expanding the traditional dynamic information system model. The new model contains difference information system and trend information system. It mines userinterested dynamic decision rules by using rough set approach and concept lattice method.The main work in this thesis includes the following aspects:1. Dynamic information system transformation model and formal description of thetransformation rules are proposed. We extend dynamic information system, and give dynamic information system transformation model based on the observation spot. Furthermore, we construct the class partition mechanism and the relevant semantics of information changing trend, while studying the relevance between condition attribute transition and decision attribute transition, and we provide the formalization of trend rule. This work extends traditional classification methods and its application model in the static information system, which play a role in the dynamic decision information system based on time sequence, so transformation rules can be mined.2. We give the definition of difference information system, and then propose a algorithm for decision rules mining——DI_FindRules.The difference information system is composed of two decision tables ,which are in the different time points. The difference information system describes the relationship between conditional attributes' variable quantity and decision attributes' change trend. A heuristic algorithm for decision rules mining on difference information system is proposed. DI_FindRules algorithm is more effective than traditional rough set decision rules mining algorithm, because the heuristic algorithm tries to generate the most important reduction, not all possible reduction.3. We construct the object (set) trend decision table, and then raise the concept of trend concept lattice. Trend information system (or trend decision table) is constructed, according to a particular object (set) in the time sequence. Trend information system describes a single object's (set's) attribute value change trend. We extend the concept lattice theory, propose the concept of trend concept lattice, and give effective concept lattice construction algorithm (CreateDecisionLattice).We also gives decision rules mining algorithm (FindRules_DecisionLattice). Since the purpose of constructing the trend concept lattice is to extract trend-related decision rules, the new lattice construction algorithm makes an improvement to omit all concepts in the lattice, which are irrelevant to the decision-making. The new algorithm not only reduces the number of concepts, improves the efficiency of rules mining under the premise of extracting the same decision rules.4. We analyze the problems by using differences information systems in practical application. e.g.: weak in anti-noise, too may rules, many new inconsistencies. We propose a new algorithm——RSTT to extract decision rules ,which meet the support and confidence level threshold.5. We begin with the data from stock transactions, then extract multi-properties, which have significant impact on the stock price and may reflect the company's financial situation, to construct stock transactions decision table. Decision tables in several time points may constitute a sequence of dynamic information system, so we can carry on exploratory research to dynamic information system, by using dynamic information system decision rules extraction model. Through comparing with the Lindig algorithm, lattice construction algorithm—Create_DecisionLattice, decision rules mining algorithm—FindRules_DecisionLattice are more efficient, and mining rules are more concise and useful.