Finding the Optimal Features Reduct, a Hybrid Model of Rough Set and Polar Bear Optimization

摘要

The target of this research is reducing the size of a dataset which is usually needed before starting the data analysis in scientific research, this can be done by removing attributes that do not affect the accuracy of the dataset, this process will enhance the performance of data analysis and will result in more accurate results and decisions. The rough set theory introduced by Pawlak provides a powerful technique to measure the influence of each attribute in the dataset and the effect of excluding an attribute on the accuracy of the dataset. However, evaluating all possible combinations of features is an NP-Hard problem and is usually not possible when dealing with datasets having large number of attributes. To solve this kind of problems the heuristics algorithms can play a crucial role to avoid scanning all possible combinations. Polar Bear Optimization Algorithm PBO, is a pretty new meta heuristic algorithm has an advantage over other heuristic algorithms of solving such kind of problems using dynamic population with a flexible production and death mechanism, which results in finding optimal solution quickly by keep producing good solutions out of hopeful candidates and keep removing unpromising ones. Our proposed algorithm could find the optimal reduct in better performance comparing to other algorithms in terms of execution time, population size and number of iterations.