Enhanced cultural algorithm to solve multi-objective attribute reduction based on rough set theory

In extracting hidden information from a data, its high dimension can create challenges in the quality of the extracted information and the search space size. Attribute reduction based on minimizing both missed information and selected subset attributes is logical solution for the challenge. Rough set theory (RST) is an information recognition technique in uncertain data that it shows the value missed information for the selected attributes. In this paper, a multi-objective attribute reduction (MOAR) is modeled by designing a new effective cost function to optimize the minimum number of attributes with the maximum dependency coefficient of the RST. Due to the MOAR is an NP-hard problem, an enhanced draft of cultural algorithm, as a continuous optimization algorithm, is proposed to solve it, as a discrete problem for the first time. The cultural algorithm (CA) with a dual inheritance system is enhanced by utilizing just normative and situational components to generate new individuals and planning a novel heuristic to discrete population and belief spaces. With regard to design the research problem, the CA and five algorithms are implemented to compare their results on twelve well-known UCI datasets in three categories sizes; small, middle and large. The tuning algorithm's parameters to find the best possible values are done and different size of the population is considered to evaluate the sensitivity of the algorithms on the population size parameter. The experimental results show that the proposed algorithm is able to find competitive results when compared to the state-of-the-art algorithms.