I am using the model to see the potential distribution for plant species.
Is there any technique to validate the output or check the accuracy?
Thank you
You should be aware of how are all complexity measures based on the application of Shannon entropy designed. Once you know this, you are the king. You become the master of all complexity measures. :-)
You have the original system expressing complex behavior. You have not the knowledge of its internal structure and function.
You measure some quantity of the systems topology, biosignal, response, etc. This measure is inserted into Shannon entropy.
From that evaluated entropy, you can deduce something about the state of the original -- not well understood & complicated -- system.
When you want to know more about this procedure read my preprint dealing with the prediction of arrhythmias from one lead ECG recordings, which is nothing else than comprimation of the complex system -- cardiovascular system including regulation parts -- into one signal (aka comprimation) that is studied using permutation entropy.
https://www.researchgate.net/publication/340460356_Application_of_Machine_Learning_and_Complexity_in_Medicine_Prediction_of_Drug_Induced_Arrhythmias_in_Rabbit_Model_up_to_One_Hour_Before_Their_Onset_early_version_requested_due_to_COVID-19_unfinished_m
Check the final publication in the future as it has extended the entropy section (which describes all details, not included in the preprint COVID-19 version) and rewritten methods section with all ML methods.
The same general methodology can be applied in your case. It works not only for temporal signals but even for topological information as in your case.
One important remark. It is possible to use cellular automata to measure spatial distributions in automatic mode. It all depends on how data are collected. You can check the review on CA in my profile.
It is imaginable that when data are collected in some sort of automatic mode, a CA can discern distributions in spatial dimensions.
The principle of maximum entropy states that the most characteristic distributions of the probabilities of states of an indefinite environment are those distributions that maximize the selected measure of uncertainty for given information about the "behavior" of the environment. For the first time, D. Gibbs used a similar approach to find extremal distribution functions of physical ensembles of particles. Subsequently, E. Jaines proposed a formalism for reconstructing the unknown laws of the distribution of random variables in the presence of restrictions from the conditions of Shannon's maximum entropy. In addition, the Shannon maximum entropy and the Lagrange multiplier method must be taken into account. In addition, http://www.cs.cmu.edu/~./aberger/maxent.html is described in detail here. There is an article https://onlinelibrary.wiley.com/doi/full/10.1111/j.0906-7590.2008.5203.s and
http://www.mathnet.ru/links/390ca9f3c228511b07ae6ee73198620c/trspy733.pdf
Regards, Sergey Viktorovich Pushkin
Masoom Reza Most of the researchers are using AUC to validate the output of Maxent. If the AUC of the model is 0.5 it is considered random, 0.75 it is fair, above 0.75 it is good and above 0.9 it is excellent. However, some researchers have pointed out that AUC is not only the basis of validating the results and they suggest Kappa statistics and True Skill Statistic as better methods to validate the performance of model. Recently, I found some researchers have suggested True Skill Statistic as the best method to validate the model of maxent. True Skill Statistic (TSS) is nothing but the sensitivity + specificity - 1. Someone has given the procedure of TSS calculation in MSExcel which if you want I can share you.
- Badges
- Science topic
- Similar topics
- Counseling Psychology
- Confusion