IMPROVED PSO BASED CLASSIFIER FOR MULTICLASS DATASETS
Loading...
Date
item.page.authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Classification can be easily encoded as a multivariable optimization problem. When in a multidimensional
newlinespace a class prototype is represented by a centroid, classification can be seen as the problem of
newlinefinding the optimal positions of all the class centroids i.e. determining for any centroid, it s optimal coordinates.
newlinePSO is very effective in solving multivariable problems, where variables take on real values which are taken as
newlinea stand-alone technique to classify the datasets and to study the PSO based Classifier for Multiclass Data Sets.
newlineThis research work therefore contributes towards improving classifiers performance by using metaheuristics
newlinetechniques.
newlineThe rise of Metaheuristics can mainly be attributed to the increase in data generation. This is due to the
newlineincrease in the information leveraging devices such as sensors, high resolution cameras and video recorders,
newlinesatellites and user information generated from the internet. Meta heuristic algorithms perform effective
newlineoptimization or promises to provide near optimal results. This method is embedded in order to provide the much
newlineneeded time efficiency. PSO is a metaheuristic technique used to optimize a problem by improving the
newlinecandidate solutions in a continuous manner. The process of optimization is carried out by components called
newlineparticles. The movement of these particles in accordance to the fitness function determines the direction and
newlinevelocity of movement of the particles. Feature subset selection is the process of identifying and removing as
newlinemuch of the irrelevant and redundant information as possible. PSO is applied on the pruned dataset to produce
newlineefficient results.
newlineIn the first stage, modified PSO algorithm (MPSO) is integrated with attribute elimination techniques to
newlineincrease the accuracy of classifier. The process begins with the evaluation of attributes using the Correlation
newlinebased Feature Subset (CFS) evaluator. Next the Greedy Hill Climbing method is used to filter attributes and the
newlinefinal pruned dataset is created. It is embedded with PSO to cre