http://www.dbs.ifi.lmu.de/Publikationen/Papers/LOF.pdf

1.

I don't understand of k-distance(o) inside reach-distr definition reach-distk(p, o) = max { k-distance(o), d(p, o) }?

2.

in my opinion before i use lof,i have to scale(scale function in R) and then calculate dissimilarity matrix(dist or daisy functions in R). what do you think?

3.

the algorithm know to work with category/binary variables, but such variables worse result, i.e i have to calculate lof do in every cluster(binary variables). for example : i want to find bond outliers in every credit(AAA,AA+...). there are a few problems with it, some of bonds have a Securities(binary variable : yes=1,no=0) and every one has a level of Seniority(category variable).

how can i take this information in to consideration, in addition to yield and duration of bond? i ran algorithm and got that every bonds that have a security are outliers and it is not right.