Machine Learning - Applications

Synthetic Minority Over-sampling Technique

Regular SMOTE

$s_i=x_i+(x_{zi}-x_i)\times\lambda,\ \lambda \in [0,1]$

Borderline SMOTE

$s_i=x_i+(p_i^{'}-x_i)\times\lambda,\ \lambda \in [0,1]$

SVM SMOTE

$s_i=sv_i+(sv_i-x_{im})\times\lambda,\ \lambda \in [0,1]$

$s_i=sv_i+(x_{im}-sv_i)\times\lambda,\ \lambda \in [0,1]$

Adaptive Synthetic Sampling Approach for Imbalanced Learning

same as SMOTE just with a minor improvement
adds a random small values to the points thus making it have a little more variance
uilizes the density distribution of a minority sample as criteria for how much synthetic data should be created.
helps focus attention on minority examples depending on how difficult they are to learn

$r_i=\frac{\Delta_i}K,\ i=1,2,...m_s$

$\hat{r_i}=r_i/\sum_{i=1}^{m_s}r_i$

$g_i=\hat{r_i}\times G$

Up Sampling