Classification Models with Copula Functions.

MLCOPULA

This package implements 7 copulas for supervised classification: frank, gaussian, clayton, joe, gumbel, AMH and grid. The classification model is based on the Bayes theorem, similar to the naive Bayes classifier model, but does not assume that the features are independent.

The probability of a class given a set of characteristics (predictor variables) is:

$$P(A|x_1,..x_d) \alpha \prod_{i = 1}^{d}f_{X_i|A}(x_i)c(u_1,...,u_i)$$

where each $u_i = F_{X_i|A}(x_i)$ with $i = 1,2,..d$.

The copula density function $c(u_1,..u_i)$ is modeled by bivariate copula functions, using graphical models (trees and chains)

Copulas

For more details and visualization of the copulas, go to https://abrahammontoyacalzada.shinyapps.io/Copulas/

Frank copula:

$$C(u_1,u_2;\theta) = -\frac{1}{\theta} ln \left[ 1 + \frac{(e^{-\theta u_1} - 1) (e^{-\theta u_2} - 1) } {e^{-\theta} - 1} \right]$$

with $\theta \in (-\infty,\infty)/0$

This copula has no upper nor lower tail dependency.

Clayton copula:

$$C(u_1,u_2;\theta) = \left( u_1^{-\theta} + u_2^{-\theta} - 1 \right)^{-1/\theta}$$

with $\theta \in [-1,\infty)/0$

When $\theta \geq 0$ has lower tail dependence equal to $\lambda_L = 2^{-1/\theta}$

Gaussiana (Normal) copula $$C(u_1,u_2;\theta) = \Phi_G (\Phi^{-1} (u_1) , \Phi^{-1} (u_2) )$$

with $\theta \in (-1,1)$

This copula has no upper nor lower tail dependency.

Joe copula $$C(u_1,u_2) = 1 - \left[ (1 - u_1)^\theta + (1 - u_2)^\theta - (1 - u_1)^\theta (1 - u_2)^\theta \right ] ^ {1/\theta}$$

with $\theta \in [1,\infty)$

This copula has upper tail dependence equal to $\lambda_U = 2 - 2^{1/\theta}$

Gumbel copula

$$C(u_1,u_2) = exp \left[ - \left[ ( -ln(u_1) )^\theta + ( -ln(u_2) )^\theta \right]^{1/\theta} \right]$$

with $\theta \in [1,\infty)$

This copula has upper tail dependence equal to $\lambda_U = 2 - 2^{1/\theta}$

Ali–Mikhail–Haq copula

$$C(u_1,u_2) = \frac{u_1 u_2}{1 - \theta (1 - u_1)(1- u_2)}$$

with $\theta \in [-1,1)$

This copula has no upper nor lower tail dependency.

Example for iris data set

library(MLCOPULA)
X <- iris[,1:4]
y <- iris$Species
model <- copulaClassifier(X = X, y = y, copula = "frank",
                      distribution = "kernel", graph_model = "tree")
y_pred <- copulaPredict(X = X, model = model)
table(y,y_pred$class)
#Example 2
X <- iris[,1:4]
y <- iris$Species
model <- copulaClassifier(X = X, y = y, copula = c("frank","clayton"), 
                        distribution = "kernel", graph_model = "chain")
y_pred <- copulaPredict(X = X, model = model)
table(y,y_pred$class)

References

Salinas-Gutiérrez, R., Hernández-Aguirre, A., Villa-Diharce, E.R. (2014). Copula selection for graphical models in continuous Estimation of Distribution Algorithms. Computational Statistics, 29(3–4):685–713. https://doi.org/10.1007/s00180-013-0457-y.