Comparison of Machine Learning Methods with Clear and Unclear Model Interpretation

Različiti algoritmi klasifikacije, osim po samom pristupu problemu i performansama, razlikuju se i po razini razumljivosti i jednostavnosti objašnjenja čovjeku. Oni se mogu formalno usporediti po preciznosti kvalifikacije nad skupovima podataka koristeći metode poput unakrsne validacije, dok za interpretabilnost formalna definicija ne postoji. Ipak, spo- sobnost shvaćanja kako algoritmi donose odluke je sve bitniji s porastom odgovornosti sus- tava koji te algoritme koriste. Iako je generalan konsenzus da kompleksniji i teže razumljivi algoritmi donose bolje rezultate, oni jednostavniji imaju svoje prednosti i primjenjivi su na odred̄ene probleme.Different classification algorithms differ by their problem approach and performances, but also by their level of interpretability and ease of explanation. They can formally be compared using various methods such as cross validation, while no definition for interpreta- bility exists. However, the ability to clearly explain how an algorithm works is growing in importance with the rapid growth of the algorithms’ responsibilities. Although the general concensus is that more complex algorithms perform better, there are cases where the simpler models can prove more capable