Research Article
A Classifier Graph Based Recurring Concept Detection and Prediction Approach
Algorithm 2: Recurrent Detection and Prediction | Input: S: data stream, max: the maximum number of nodes in G; | Output: G; | begin | create a graph G= Null; | for each instance ins in S do | DBDM(ins); | if current state is Warning then | store ins in Bn; | train Cn for later use; | end if | else if current state is Drift then | for each arrow in G whose from-node is p | choose the arrow which from-node is Vk with the maximum weight; | if compare (Bn, Vk.instances) then | Cn = Vk.C; | clear(Bn); | else | create a new node to store Bn and Cn; | if vexnum > max then | delete one node in G; | insert new node into G; | Cn replace the current classifier; | end if | end if | end for | else | clear Bn; | end if | end for | return G; | end. |
|