2
$\begingroup$

Context: Fitting a Machine Learning Algorithm on a labeled dataset.

For a feature vector [a,b,c] and a labeled output/target variable, what's the term used when identical feature vectors map to two (or more) different target variables? I've been using 'coincident feature target collision' but I'm sure there's a better, more formal, community accepted term for this. :)

import pandas as pd
data = {'a':[0,0,0,0,0], 'b':[0,0,1,1,1], 'c':[0,1,0,1,0], 'target':[0,1,0,1,1]}
df = pd.DataFrame(data)
print(df.head())

   a  b  c  target
0  0  0  0       0
1  0  0  1       1
2  0  1  0       0   <--- feature vector (0,1,0)  Target = 0
3  0  1  1       1
4  0  1  0       1   <--- feature vector (0,1,0)  Target = 1

In the machine learning context, we observe that for this scenario the machine learning algorithm will not be able to properly fit a model that predicts the target variable.

Update: @whuber thanks for the comment. Looking around a bit I see you also made a comment on this somewhat related question: reducible vs irreducible error.

Also from this article the term 'irreducible error' is called out https://machinelearningmastery.com/gentle-introduction-to-the-bias-variance-trade-off-in-machine-learning/.

"The irreducible error cannot be reduced regardless of what algorithm is used"

That article also gives this definition

"Variance is the amount that the estimate of the target function will change if different training data was used."

So using the general term 'variability' for this particular issue which I'm clumsily calling 'coincident feature target collision' would be ambiguous at best and highly confusing at worst. I feel like it's part of the 'irreducible error'. Another quote from the article:

It is the error introduced from the chosen framing of the problem and may be caused by factors like unknown variables that influence the mapping of the input variables to the output variable.

$\endgroup$
  • 2
    $\begingroup$ "Variability" is standard among statisticians :-). $\endgroup$ – whuber Mar 4 at 22:50
  • 1
    $\begingroup$ In regression analysis variance based only on such cases is called pure error. Comparing this pure error variance with residual error variance (from a model) is a way to see if the used features really captures everything, or if there still is unexplained variance. See this list of relevant posts $\endgroup$ – kjetil b halvorsen Mar 5 at 17:24
1
$\begingroup$

The error that comes from having multiple identical inputs associated with different outputs is called the Bayes error or irreducible error. I don't know a general name for this type of data, but depending on why this is the case they can have different names:
- Distribution overlap
- Stochastic output
- Noise

$\endgroup$
  • $\begingroup$ Great, thanks exactly what I was looking for :) $\endgroup$ – Briford Wylie Mar 5 at 18:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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