A Semisupervised Cascade Classification Algorithm

Algorithm 2

The self-trained (NBC4.5) algorithm.

Algorithm  Self-trained (NB C4.5)

Input:

NBC4.5 – Cascading NB and C4.5, as base classifier

– Initial training dataset

– initial labeled examples,

– initial unlabeled examples,

– Examples with Most Confident Predictions

AccT – Threshold of acceptance

MaxIter – number of maximum iterations performed

    Initialization:

     Train NBC4.5 as base model on

    Loop for a number of iterations (MaxIter is equal to 40 for our implementation)

     Use NBC4.5 classifier to select the examples with Most Confident Predictions per iteration ()

     Remove from and add them to

     In each iteration a few examples per class are removed from and added to

     Re-train NBC4.5 as base model on new enlarged

Output:

Use NBC4.5 trained on to predict class labels of the test cases.