Naive Bayes Classifier for Cost Sensitive Dynamic Learning

Both cost-sensitive classification and online learning have been broadly researched in data mining and machine learning communities, respectively. However, very restricted study addresses an essential intersecting issue, that is, “Cost-Sensitive Online Classification". In this paper we proposed, formally study this issue, and propose a new framework for Cost-Sensitive Online Classification by directly surging cost-sensitive symptoms implementing online gradient descent methods. Particularly, we present two novel cost-sensitive online classification algorithms, which are structured to directly amend two well-known cost-sensitive measures: (i) maximization of weighted sum of sensitivity and specificity, and (ii) minimization of weighted misclassification cost. We study the theoretical leap of the cost-sensitive measures made by the presented algorithms, and broadly observed their factual operation on a variety of cost-sensitive online classification works. Finally, we represent the application of the presented method for solving several online problem detection works, showing that the presented method could be a highly effective and efficient tool to tackle cost-sensitive online classification works in several application domains. malicious Uniform Resource Locator (URL) detection is an essential issue in web search and mining, which plays a complex role in internet protection.