Univariate Gaussian Mixture Models

Mixture Models in R

Victor Medina

Researcher at The University of Edinburgh

gender %>% head()

  Gender   Height   Weight        BMI
1   Male 73.84702 241.8936 0.04435662
2   Male 68.78190 162.3105 0.03430822
3   Male 74.11011 212.7409 0.03873433
4   Male 71.73098 220.0425 0.04276545
5   Male 69.88180 206.3498 0.04225479
6   Male 67.25302 152.2122 0.03365316

gender %>% select(-Gender) %>% head()

    Height   Weight        BMI
1 73.84702 241.8936 0.04435662
2 68.78190 162.3105 0.03430822
3 74.11011 212.7409 0.03873433
4 71.73098 220.0425 0.04276545
5 69.88180 206.3498 0.04225479
6 67.25302 152.2122 0.03365316

Modeling with Mixture Models

Which is the suitable probability distribution?
How many sub-populations should we consider?
Which are the parameters and their estimations?

head(gender %>% select(-Gender))

    Height   Weight        BMI
1 73.84702 241.8936 0.04435662
2 68.78190 162.3105 0.03430822
3 74.11011 212.7409 0.03873433
4 71.73098 220.0425 0.04276545
5 69.88180 206.3498 0.04225479
6 67.25302 152.2122 0.03365316

head(gender %>% select(Weight))

gender %>% 
  ggplot(aes(x = Weight)) + geom_histogram(bins = 100)

Which distribution?

Histogram

Gaussian distributions

How many clusters?

Which parameters and how to estimate them?

Which parameters?

Two means
Two standard deviations
Two proportions

How to estimate them?

EM algorithm implemented in flexmix

Let's practice!

Mixture Models in R