Performance optimization for model training/application

[ChillkroeteTTS]

By using the clojure standard library for computing the anomaly detection model, many features will lead to a long training time. (1000 feature vectors with 1000 features each took 1.377s on an i5-4300U CPU @ 1.90GHz took)

I never had a closer look for areas of optimization but I'm sure there is much potential left since my implementation never had performance in mind.