0
$\begingroup$

From my learning and project experiences, it seems that most algorithms I got exposed to are basically doing classification (actually the only regression algorithm I could think of at this point is linear regression). I am not sure why this is the case, some random assumptions include

  • Classification algorithms could be extended to regression version without much efforts. Maybe there exists some mechanic way to do so, just like when people extend binary classification to multiclass version.
  • Contrary to the first point, it is actually technical to do such extension and so for beginners in machine learning, there is no need to study these technical tricks.

Could someone give me some hints about this problem, thank you in advance.

EDIT:

I think there are some issues with this question and what is actually interesting to know is that why more classification algorithms are taught in introductory or even advanced machine learning courses than regression and other algorithms.

Examples:

  • Introductory course: CM 146 offered at UCLA
  • Advanced course: CS 260 also offered at UCLA
$\endgroup$

closed as off-topic by conjectures, Peter Flom Dec 25 '18 at 12:15

  • This question does not appear to be about statistics within the scope defined in the help center.
If this question can be reworded to fit the rules in the help center, please edit the question.

  • 3
    $\begingroup$ IMO the question is too broad and primary opinion-based. Algorithms are solving problems. Therefore there are many for classification and also many for regression, clustering, etc. And btw. how you count algorithms? E.g. there are tens of variations of decision trees. $\endgroup$ – wind Dec 19 '18 at 6:48
  • $\begingroup$ @wind plz see my edit $\endgroup$ – Mr.Robot Dec 19 '18 at 12:34
  • $\begingroup$ Looking at the schedules, neither of the courses you mention needs to focus more on classification then regression. There are regression equivalents for decision trees, SVM, KNN etc., so we can assume as well that they are mentioned. $\endgroup$ – Tim Dec 19 '18 at 13:53
  • 1
    $\begingroup$ @Mr.Robot I think classification is used more often for basic courses, because the process of data cleaning, feature extractions is the same for both classification and regression, while the classification can be explained simple because of two classes on the output. Regression has continous output, so it requires more theoretical background. But it's still my opinion rather than fact. $\endgroup$ – wind Dec 19 '18 at 14:32
  • $\begingroup$ @Tim even though there does exist regression versions of decision trees, SVM, and others, those materials are not covered in both courses. In fact, neither the lectures nor the text used for the course covers these topics, probably due to the time constraint. $\endgroup$ – Mr.Robot Dec 20 '18 at 2:50
1
$\begingroup$

There are machine learning algorithms that implement 'probability machines' directly. But the overuse of algorithms that do up-front classification is an alarming trend in machine learning and is poorly thought out. I have written about this in detail here.

$\endgroup$
  • 1
    $\begingroup$ Your posts about classification are very interesting to read and they provide additional insights. Thank you for sharing :-). $\endgroup$ – Mr.Robot Dec 20 '18 at 2:59
0
$\begingroup$

What exactly do you call a "machine learning algorithm"?

If you take a reasonably broad view of what MLA means, then there are many. Since you count linear regression as an MLA, you clearly take a broad view. You then have all sorts of variations on linear regression from generalized additive models to quantile regression to multivariate adaptive regression splines (MARS).

And what about factor analysis, cluster analysis, multi-dimensional scaling and so on and on?

Of course, there are regression trees, just as there are classification trees, and each method has spawned offspring.

But, if it somehow does turn out that you are right and there are more for classification, then I think one reason might be that people often wrongly turn continuous outcomes into categorical ones.

$\endgroup$
  • $\begingroup$ I just realized the issues with my question. But by machine learning algorithm I mean the algorithms that are introduced to students in introductory machine learning course, like the one I used to take. $\endgroup$ – Mr.Robot Dec 19 '18 at 12:27

Not the answer you're looking for? Browse other questions tagged or ask your own question.