0

I have few basic doubts regarding Hypothesis testing,

I know Hypothesis Testing is a statistical test, for a sample of data stands true for the entire population or not. That is, if a random sample's mean is same as that of the population mean. Here, we try to accept or reject NULL hypothesis by various tests like Z-Test/ T-Test / ANOVA / Chi-Square Test.

  1. What we do after accepting or rejecting NULL hypothesis?
  2. Do we exclude/include that sample from further process if we are building a machine learning model?
  3. What are the significance of accepting NULL Hypothesis?
  4. What are the significance of accepting Alternate Hypothesis?
  5. Or is there any other insights we make with these tests?

I would like to know these in the perspective of machine learning for model building.

Kindly share your thoughts.

Vin
  • 89
  • 5
  • 1
    please see [stats.se] for non programming related questions. – hongsy Feb 07 '20 at 09:23
  • This question is better suited to stats.stackexchange.com. That said, there is a pretty short answer, which is that significance tests and/or hypothesis tests are a useless distraction when working with machine learning (and in other contexts as well, but you didn't ask about that). With a typical sample size in the millions, every microscopic difference is "significant". What does that mean? Who cares? The right way to go about assessing and comparing ML models is to work with practical significance as measured by out-of-sample error or a plausible stand-in such as cross validation error. – Robert Dodier Feb 07 '20 at 20:31
  • Do you mean to say there is no or minimal use of such testing in real practice? We consider the samples for ML purpose for any decision through Hypo testing? – Vin Feb 13 '20 at 06:15

0 Answers0