Main Article Content
Abstract
The cost-sensitive classification and online learning have been well considered individually in data mining and classification, there was very few wide-ranging learning of cost-sensitive online classification in review work. On the other hand, recent traditional algorithms simply focused first-order data of data stream. It is inadequate in tradition, because numerous existing methods have proved with the purpose of integrating second-order data improves the classification results of classifiers. To manage this problem, Adaptive Cost-Sensitive Online Gradient Descent (ACOG) classifier by adaptive regularization is developed recently. On the other hand in ACOG classifier, optimization of the cost function becomes extremely hard task. To handle this problem, Swallow Swarm Optimization (SSO) algorithm is introduced which optimizes the parameters of the cost for online gradient Descent classifier. Reduced error classification results parameters designed for positive and negative samples are optimized by SSO algorithm. SSO algorithm includes of three major types of particles: explorer particles, aimless particles, and leader particles. Every particle has an individual characteristic designed for optimization of the cost parameters designed for inner colony of flying. Every particle shows an intelligent behavior and, continuously, discovers its surroundings by means of a reduced error value. Subsequently designed for improved trade off among the results and effectiveness, additional develop the sketching algorithm, which considerably speed up the computation time by means of moderately small results loss. Hypothetically examine the proposed classifiers and existing algorithms in wide experiments by means of the german and covertype dataset. Classifiers are experimented in MATLAB environment and measured by means of sensitivity; specificity, sum, and computation time.