Hi all,
Can anyone provide more explanation about statistical tests.
I have obtained a new result in my application, so I need to compare my result with the original result of the Algorithm.
Which good statistical test applying to compare between two matrix.
Test do not compare anything. Test are a procedure for making decisions about accepting or rejecting a hypothesis under the control of the error-rates (false accepances and false rejections). This is not a comparison.
If you want a comparison, than think what exactly you want to compare and simply compare it - maybe give a difference or a ratio (along with confidence intervals), whatever seems apropriate to your problem, a description of more complex structures is also possible. It depends pretty much on the specific things you want to compare, and to tell you what would be good, appropriate or "the best" (there is nothing like the "best" anyway) you will have to explain every little detail of your aim, the experiment, the measurements, and the results you have. If you need good advice there, this is presumably best discussed with a local statistician.
If you want a test (not a comparison), then please specify the hypothesis you want to test exactly. This means, boil down your problem to the calculation of some (ideally sufficient) summary statistic and then it is either already clear what test can be used (e.g. to test the hypothesis that such statistics are the same for your results and the results of the algorithm) or you still need help, in which case you can ask again, but describing the precise hypothesis you want to test. If you have problems specifying the precise hypothesis, then again useful help can only be provided when all the details are clear, so in theis case again it seems best to consult a local statistician.
Test do not compare anything. Test are a procedure for making decisions about accepting or rejecting a hypothesis under the control of the error-rates (false accepances and false rejections). This is not a comparison.
If you want a comparison, than think what exactly you want to compare and simply compare it - maybe give a difference or a ratio (along with confidence intervals), whatever seems apropriate to your problem, a description of more complex structures is also possible. It depends pretty much on the specific things you want to compare, and to tell you what would be good, appropriate or "the best" (there is nothing like the "best" anyway) you will have to explain every little detail of your aim, the experiment, the measurements, and the results you have. If you need good advice there, this is presumably best discussed with a local statistician.
If you want a test (not a comparison), then please specify the hypothesis you want to test exactly. This means, boil down your problem to the calculation of some (ideally sufficient) summary statistic and then it is either already clear what test can be used (e.g. to test the hypothesis that such statistics are the same for your results and the results of the algorithm) or you still need help, in which case you can ask again, but describing the precise hypothesis you want to test. If you have problems specifying the precise hypothesis, then again useful help can only be provided when all the details are clear, so in theis case again it seems best to consult a local statistician.
T test (parametric) or test of Kappa (nonparametric).
The nonparametric test of Kappa allows to quantify the agreement between 2 estimators when it concerns a qualitative parameter. The agreement is defined as the accordance of 2 informations about a same object. This notion implies existence of a link between variables, requires same kind of variables and a matching of judgments.
So many statistical significance tests are available....to compare multiple algorithms for single dataset.. Mcnemar's test is applicable....for multiple datasets u can opt wilcoxons test..... depends on problem design and hypothesis chosen. ....
Hi;
you can test your algoithm using the confusion matrix and calculate accuracy from the matrix result
Thank you Very much All,
Statistic isn't my domaine of Research, So i havn't as sufficient information about Statistical Test...
For this reason i search one who can help me more
Thank you all again
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R Statistical Package