Feature selection is the process of choosing a subset of relevant as well as irredundant features from a bigger set. In other words, it removes redundant and irrelevant features from original set. In this paper, a new algorithm which is called bidirectional ant colony optimization feature selection (BDACOFS) based on ant colony optimization (ACO) algorithm and inspired from ACOFS (a recently proposed feature selection method) is presented. In the proposed algorithm, problem is modeled by a circular graph in which every node has only two arcs to its subsequent node. One of arcs represents selecting and another implies deselecting the next node. In addition, heuristic desirability of every node's selection is calculated according to two factors; one is related to discrimination ability of features and second one is related to mutual information among features. The proposed algorithm has been tested against some wellknown datasets and its performance has been compared to some well-known algorithms. The result indicates that proposed algorithm by adding mutual statistical information to its heuristic desirability could remove more redundant features than original ACOFS. Meanwhile it keeps classification accuracy as highly as the original ACOFS.
Feature selection is a process with the aim of diminishing irrelevant and redundant features from original set. Some of features which are called irrelevant not only provide no useful information to classifier but also put it in pitfalls. Some other are neutral means no conducting and no misleading. This group is called redundant features. In this paper, we introduced a new algorithm based on ant colony optimization (ACO) with the aid of statistical information from training dataset to cut down mentioned groups of features. The proposed algorithm is called Sequence Based Feature Selection (SFS) able to get target subset cardinality as an input or not. In this paper ants are applying two factors related to statistical information as heuristic desirability to remove irrelevant and redundant groups of features. These information are well-known Fisher-Score and Correlation respectively. Our algorithm utilizes a fully connected graph of nodes to model the problem and put the pheromone on nodes rather than edges. In addition it uses a dynamic desirability rather than a pre-calculated static one. In each iteration every ant comes up with a sequence of features which is passed to the next step for choosing a subsequence of it starting from beginning. The proposed algorithm has been tested against some well-known datasets and its peiformance compared with recently represented algorithms. The results indicates that using dynamic two factors desirability in fUlly connected graph is more effective in removing redundant features than Binary approaches.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.