Suppose that i want to compare the accuracy of prediction between A and B. The number pf experiment actually is more than 100 experiments. I only put 6 experiments for example. If i use the independent t-test to compare between 2 groups, the results only indicate that the means between two groups is different or not but fail to answer that which group is more superior in term of accuracy (it is bias if we only compare them in term of mean only.."higher mean means that that group is more superior")..if i calculate the difference between the group (+ve if the left hand is higher value and -ve if right hand is higher value) and create a bar chart to see which group is superior, i found out that it is not good for academic writing.
The Receiver Operating Characteristic (ROC) curve is the easy way to understand and makes comparison of the performance of methods.
But the comparison of means is a test of which one is more accurate since your variable is 'accuracy'. But you need to do a chi square test comparing frequency distributions
Hi, Ahmad,
Your question asks about the identity of the best test to compare the accuracy of prediction between A and B.
I am able to definitively state which method ("test") identifies the model which maximizes the accuracy in predicting between A and B.
Here is a 3-page illustrated article about the only method which is guaranteed to identify the most accurate model which exists for a sample given the data and hypothesis.
- https://odajournal.com/2018/11/26/visualizing-application-and-summarizing-accuracy-of-oda-models/
And here is a brief review of the maximum-accuracy statistical paradigm.
- https://odajournal.com/2020/06/28/what-is-novometric-data-analysis/
- Badges
- Science topic
- Similar topics
- Education
- Higher Education
- Academic Writing