I want to know different methods or techniques by which I can get to know exact parameters to be considered while performing sensitivity and stability analysis for a biological model.
Danish -
You might start by coming up with a list of variables for which you have good quality data, that you might think could be predictors for your variable of interest. Then try some regressions and estimate the variance of the prediction error in various cases. You may need multiple nonlinear regression, but I think you will want to see what might be the simplest model you can find which performs well, to avoid over-parameterizing, which may be misleading.
Looking at scatterplots can be informative. For multiple regression, you can plot the 'predicted' (estimated) y value on the x-axis, and the y (variable of interest) value on the y-axis. That could be informative.
There are techniques (backward and forward elimination could be searched on the Internet) for reducing the number of regressors, but as was discussed in answers to the question at the link attached, see Theo Dijkstra's comments in particular, you should have good reasons for using the predictors that you use. Some may be 'discovered' through statistical experimentation, but may turn out to be spurious, so you will need scientific investigation/knowledge from your field, as well as statistical science, to find a good model or models.
There are various interactions between regressors that can occur, so you will want to be aware of that.
To explore a bit, you might want to first look at x-y scatterplots using one predictor at a time.
I hope your project is enjoyable. - Jim
https://www.researchgate.net/post/When_doing_multiple_regression_analysis_should_I_put_all_predictors_in_regression_model_or_only_those_having_a_significant_correlation_with_outcome
Hi Danish,
you might face problem of "local minimum/maximum point" in the parameter space. There are no general method to obtain global optimum in iterative methods....One simple way to improve the result of parameters estimation is random choice of starting points for iterations in the parameter space,,,and then making comparison of the functional values...I'm sure you can find some software where you can perform this automatically ...
Vlad
- Badges
- Science topic
- Similar topics
- Engineering
- Control Theory
- Control Systems
- Control Systems Engineering