Description
Background
The benchmarks of anomaly algorithms are important to determine the best algorithm given a dataset. The discussion in my Sklearn's PR#16378 shifted the focus from creating a quantitative benchmarks for anomaly algorithms to creating ROC examples of Local Outlier Factor (LOF) and Isolation Forest (IF). The suggestion is in the last comment of Sklearn's PR#9798
Challenges
- LOF didn't originally create for outlier detection context (training set = testing set). Thus, it doesn't have a
decision_functionto compute ROC curve
Plans
- Apply an algorithm in sklearn PR#9798 to create ROC in LOF
- Create ROC curves from algorithm LOF and IF, using datasets from
sklearn.dataset - After peer review and TA review, PR into "scikit-learn/scikit-learn/benchmarks"