Survey of Novel Method for Online Classification in Data Mining

Nowadays in communities of Data Mining and Machine Learning, cost-sensitive classification and online learning have been widely examined. Even though these topics are getting more and more attention, very few studies are based on an important concern of Cost-Sensitive Online Classification. This problem can be explored widely and new technique can be implemented to deal with this issue. By directly enhancing cost-sensitive measures utilizing online gradient descent methods, a new technique can be implemented. Particularly, two novel cost-sensitive online classification algorithms can be presented, which are intended to specifically enhance two well-known cost-sensitive measures: (a) maximizing weighted sum of specificity and sensitivity, (b) minimizing weighted misclassification cost. The hypothetical limits of the cost-sensitive measures made by the algorithms should be examined. Also their experimental performance on number of different cost-sensitive online classification tasks should be examined. This technique can be efficiently utilized for solving number of online anomaly detection tasks. Also it is very efficient and effective technique to handle cost-sensitive online classification tasks in number of application domains.