A Hybrid Sampling SVM Approach to Imbalanced Data Classification

Algorithm 1

The hybrid sampling SVM method.

Input: Imbalanced training data set 𝑇₀

Output: An SVM classifier 𝐹_𝑘

Step  1. Train an SVM classifier for training dataset 𝑇₀ and delete some negative samples using (5).

Step  2. Divide randomly T into k disjoint equivalent subsets 𝑇₁,𝑇₂,…,𝑇_𝑘.

Step  3. Select subset 𝑇₁ and over-sample the positive samples using SMOTE method; generate a

 new training data set 𝐺₁ by merging the new synthetic samples into 𝑇₁; train an initial SVM

 classifier 𝐹₁ for data set 𝐺₁.

Step  4. For each subset 𝑇_𝑖 in the rest 𝑘−1 subsets do

 Step  2.1. Compute the distances between negative samples and the hyperplane of classifier 𝐹_𝑖−1

  according to (5).

 Step  2.2. Select 𝑝_𝑖 negative samples with the smallest distances; generate synthetic instances

  using SMOTE method.

 Step  2.3. Merge all 𝑛_𝑖 positive samples and the new synthetic samples into 𝐺_𝑖−1, and obtain data

  set 𝐺_𝑖.

 Step  2.4. Train an SVM classifier 𝐹_𝑖 for dataset 𝐺_𝑖.

Step  5. Classify data set 𝐺_𝑘 using SVM method, and obtain a classifier 𝐹_𝑘.