0

I have an assignment of variable selection in survival analysis which I have done and on the basis of IBS (Integrated Brier Scores) I have concluded my proposed technique works well than the previous (LASSO and RSF) methods. Now I am asked to include some plots to show our approach works better than others. What are the best ways to graphically show my results plz?

I have included only the following self-explanatory plot. enter image description here Now, I want to include some boxplots or some other graphical ways to show my method works better. Please let me know what might be the best possible ways.

Basheer Ahmad
  • 1
  • What is the basis for your saying "I have concluded my proposed technique works well than the previous (LASSO and RSF) methods"? Was there some bootstrap or cross-validation involved, or are these just the Brier scores bases on single fits to each data set? – EdM May 17 '23 at 14:48
  • @EdM, I just calculated IBS values on single fits and got lower IBS for selected models. Do you know plots which can explain my scenario? – Basheer Ahmad May 17 '23 at 15:48
  • Such plots would be misleading. There's a big risk of error when you proceed this way. You might get an extremely good fit on your particular data sample, but in a way that over-fits peculiarities of your data and doesn't work well at all on new data. See this page among many others on this site. You need to validate your model-building process in some way, via things like modeling on repeated bootstrap samples or repeated cross-validation. Harrell's Regression Modeling Strategies has much useful guidance. – EdM May 17 '23 at 16:14
  • Dear @EdM, Is it okay to decide on the basis of single IBS values rather than using cross-validation or bootstrapping? – Basheer Ahmad May 18 '23 at 16:38
  • In general, no. The risk with variable-selection techniques is that they often don't extend well to new data. Bootstrapping is similar to taking many independent samples of data, so success of your new technique on multiple bootstrapped samples would support its general applicability to survival analysis. – EdM May 18 '23 at 18:36
  • Section 4.3 of Regression Modeling Strategies shows the types of problems with variable selection if you don't validate properly. If you want to show that your method works better than LASSO and random survival forests, you could make that comparison on multiple bootstrapped samples analyzed all 3 ways. – EdM May 18 '23 at 18:49

0 Answers0