Research Article
Network Traffic Classification Based on SD Sampling and Hierarchical Ensemble Learning
| | Input: Imbalanced train set , scaling factor , instance hardness threshold , and sample threshold UB | | | Output: New train set | | (1) | Step1: Distinguish between easy sets and difficult sets for each sampledo | | (2) | Compute its nearest neighbors and ifthen | | (3) | Put the samples into the difficult set | | (4) | end | | (5) | end | | (6) | Difficult set and easy set | | (7) | Step2: Compress the majority samples in the difficult set by the cluster centroid | | (8) | Take all the majority samples from and set it as | | (9) | Use the K-means algorithm with cluster | | (10) | Use the coordinates of cluster centroids and replace the majority samples in | | (11) | Compressed the majority sample set | | (12) | Step3: Sample the minority samples in the difficult set using the SMOTE algorithm | | (13) | Take all the majority samples from and set it as | | (14) | for each sample do | | (15) | Using SMOTE sampling, the sampling threshold is set to | | (16) | Putting new samples into | | (17) | end | | (18) | Step4: Merge sample sets | | (19) | New train set |
|
