Running time of Classifiers

Muhammed_Fatih_

Hello together, 

is there somebody who is able to arrange classification models according to their order of running time from quickest to slowest? I heared that Naive Bayes is the quickest one. Which one is the next quickest considering SVM, DT, Logistic Regression, kNN? 

Thanks in advance for your help! 

Best regards, 

Fatih

BalazsBarany

Hi,

your case sounds like text classification. Support vector machines are popular for this use case. But you should test some of the main algorithms regardless. In the end, run time is important, but if you can get better classification from a slower algorithm, that could be more important.

Regards,

Balázs 

BalazsBarany

Hi Fatih,

this is not possible to answer in general terms.

Some algorithms become slower when you have more data, some when you have more attributes. Sometimes they can handle nominal data faster, sometimes numerical data. For many algorithms the runtime is highly affected by parameter settings: SVM/C, Deep Learning/network structure etc.

In the case of lazy algorithms like k-NN the "learning" is very fast (just create a copy of the data), but model application is slow. 

And so on.

Model building time can be a factor in the modeling algorithm selection, but it's seldom the most important one.

You should check the running time on your own data set, that'll give you a usable answer.

Regards,

Balázs

Muhammed_Fatih_

Hi Balazs, 

thank you for your answer!

What kind of classifiers are the next quickest (after Naive Bayes) regarding the running time by taking into consideration a high number of dimensionality and rows with numerical values (TF-IDF values)? 

Best regards, 

Fatih