Hello every one,
I have 9 algorithms and 4 benchmarks, I only have the mean and standard deviation over all runs (the same number of runs), but I could not get them from the authors.
Could anyone help me in choosing the best statistical test for:
1. Multiple comparison of all the 9 algorithms
2. Pairwise comparisons of my algorithm with each of the 8 remaining algorithms
The numerical results I am aiming at comparing (and so analysing) are attached
Thank you in advance
I would not insist on doing any statistical analysis of published results, especially when the authors seem not to be open to such an analysis by you.
Addressing your question here, I would keep things simple. You can pool the sample variances to get an estimate of the random variability. Use Dunnett's test in which your algorithm is the control group.
A basic Pareto Plot of the means, standard deviations, and coefficients of variation may be useful too.
- Badges
- Science topic
- Similar topics
- Biological Science
- Microbiology
- Virology
- Virus