Create Exercise "Outlier"

[melsigl]

No description provided.

[melsigl]

There will be no outlier exercise for the summer semester of 2022, instead, we agreed upon extending the classification exercise to now comprise three weeks with the task of implementing one algorithm per week.

If we still want to create and envision an outlier exercise, I will keep this issue open.

[melsigl]

Potential exercise task ideas include:

• Parametric outlier detection based on normal distribution (slide 18)
• Parametric implementation of Mahalanobis distance (slide 21, this maybe even from scratch, although I am not sure yet).
• Non-parametric method using histogram and Kernel density estimation (slides 23 - 24)
• Distance-based outlier detection with a nested loop (slides 26-27)

Page numbers refer to the lecture slides from 2023SS (c.f. commit cab932b).

[Lucew]

I took on the following tasks:

• Findings and appropriate probability density function, where the parameter can be estimated using Maximum-Likelihood
• Use the distribution to make outlier detection
• Implement both as a jupyther notebook for the exercise