Multi-objective Algorithm And Its Application Subgroup Discovery

Subgroup discovery is a data mining technique aimed at extracting interesting rules with respect to a target value. The rules are called subgroups. This paper analyzes three subgroup discovery algorithms—CN2-SD, SDIGA and NMEEF-SD, summarizes the advantage and disadvantage of these algorithms, combines the advantage of CN2-SD and SDIGA, proposes some measures to improve the performance of NMEEF-SD. Specific research contents are as follows:Firstly, this paper introduces the definition of subgroup discovery, analyzes the research status of the subgroup discovery algorithm and main elements in subgroup discovery algorithm.Secondly, this paper analyzes the CN2-SD algorithm and investigates how to adapt classification rule learning approaches to subgroup discovery. Experimental evaluation of CN2-SD on data set shows substantial reduction of the number of induced rules, increased rule coverage and rule significance.Thirdly, a single-object GFS within the iterative rule learning approach for subgroup discovery is presented, the subgroup discovery iterative genetic algorithm, SDIGA, which obtains fuzzy rules for subgroup discovery in disjunctive normal form.Finally, a new multi-objective evolutionary algorithm for subgroup discovery with fuzzy rules is presented in this paper. The NMEEF-SD algorithm is based on the NSGA-II algorithm for the induction of rules which describe subgroups. This paper suggests some improvements for the NMEEF-SD algorithm. These improvements include introducing the concept of the weighted coverage, taking the weighted coverage as one of the targets in non-dominated sort, using the weighted coverage to select individuals from best front for crossover, introducing post-processing step to the algorithm. In the end, some experiments are performed to verify the effective of the improvements.