I participated in a contest a while ago and the goal was to determine the cause of an effect from a training set of causes and effects and finding the regularity between them and then using it to classify the test data.
I wonder if it is the same as a usual classification task or it is different somehow ?
Thanks in advance.
Hossein. It is exactly the same issue. Causes are the independent variables of any problem. Effects are the dependent variables of what you consider a problem. Let say that we want to predict/explain the temperature of an engine (no matter it is represented by a discrete value or real value), in terms of some independent variables/causes: level of cooling liquid, tire pressure, level of engine oil, current speed...hope this help...!
.
beware of mistaking correlations for causality !
most statistical analysis will detect correlations between the features and the label ; label can often be taken as an effect to be analyzed but features, even those taking part in a successful model, should not be taken as "causes" without a thorough examination
this had been discussed a while ago on RG (maybe it is about the same challenge) :
https://www.researchgate.net/post/Cause-effect_dilemma2?_sg=oXOR6nULg80j6E0t0i%2BOKzlNkZhaIJ9kzhdcr1szLVC%2FSOthwZ04eqnAdP2ZDY%2Bq_gvJKV229e8u40e9ZD1jFERfS5foLksWEPOcYlsaTN%2B2Ht%2BnrTQECHAr%2FllHye1YY_2lEoyK%2BzN8M9Kz9vB1Ui2ZQF6kD9ntRGnLQkL5ZR%2BtLe9lNEIcC0vgPTPLc95Czr#view=5223792ad4c118ed40c4ee56
.
Causes and effects are concepts that go deep in the philosophy of science.
In your practical case, I think the problem focuses on extracting patterns of occurrences. I understand that the assumption that some facts are "causes" and some facts are "effect" was given. This assumption could derive e.g. from the fact that causes may be preceding effects in time. So if A always occurs before B, i.e. there is no A which is not followed by B, they A and B can be considered cause-effect. That is in my opinion all that matters.
Complication arises when. e.g. if A then B with probability 0.9, which means that A is very likely a cause, but not alone and that another variable is involved in the process, e.g. if A & C then B with probability 1.
So eventually you have a classification problem, but augmented with a definition of "cause". Thus, a definition could look like "A is the cause of B if and only if each time A occurs, B follows with probability 1". You could have similar/slightly different definitions, and then off you go with classification.
Hume and Kant were reluctant to accept the cause/effect explanation because philosophically speaking, accepting this would lead them and others to accept the existence of a starting cause which many would associate with God. That's what I think.
Although it is indeed a controversial aspect, there are recently some interesting developments based on graphical models (see the work of J. Pearl). In principle you can distinguish correlation and causal relation by doing interventions, that is controlled experiments.
- Badges
- Science topic