As in other cases of regression, r-squared must be > 0.8. The best model is the one that can predict best your experimental results, where the difference between predicted y real results at each time be the minimum possible.

I agree with Prof. Luis Estrada-Flores, Rsquare value must be higher then 0.8. But However even If R square value less then 0.8 then still you can fit any respective model for kinetic study. Only the things you can consider while interpreting results like "Its best fit kinetic model amongst all___, ____,_____,____and ____kinetic model.

Dear Wen,

First of all a small correction to the description supporting your question. You actually fitted the models (first order and Korsmeyer-Peppas) to your experimental data, not the data to the models.

Regarding the question, I am afraid I do not agree with the previous contributions, at least without additional justification and IMHO there is no straight answer to that.

Is like asking how fast should a car be? It depends on several aspects namely on what you want the car for; in what kind of road is the car supposed to be used, etc.

I guess that you want to describe the experimental data, not to choose a model intended to predict (future) dissolution behaviour/kinetics when varying some (unknown) variables. In a general sense, the R2 tells you the amount of the overall observed data variability/change that is explained by the model. As such, a R2=0.80 means that 80% of the observed change in data is explained by the underlying model. It does not allows you to compare different models and say that one is better than the other or to make any conclusion about the model validity.

In the case of the R2=0.80, what are the remaining 20% of the experimental data change (not explained by the model) due to? To experimental variability? To the model inability to describe properly the data (lack-of-fit)?

Without further information I’m not able to give an answer stating the 0.85 or 0.99 are OK or not (despite the fact that I found 0.99 exceptionally high).

Are you able, without disclosing unnecessary information, to share the experimental data you fitted the models to, as well as the resulting model parameters and results? (preferably in a general format as, for instance, “csv”, “txt”, Excel). Making the data available will help me (and other fellows) to give you an more accurate and helpful answer.

Regards, Luis

PS: Dear Luis and Vipul, would you be so kind to give a more detailed support to your answers and help me to understand them? “As in other cases of regression, r-squared must be > 0.8.”. As I’m not acquainted to that rule, could you please give me a bibliographic reference I could read? Could you please send a bibliographic reference to support “The best model is the one that can predict best your experimental results” (and/or explain why)? Thanks in advance and regards.

Dear Wen,

Now, with the help of the additional info, it's more clear what you intend to do/conclude with the model fitting.

I just took a quick look at your data. Are you able to include the raw data (I mean,  the dissolved amount vs time for the 4 formulations, including the replicates which I guess are available)? and then let us perform the fitting and interpret the results?

There is no better diagnostic tool that plotting the data and look at the resulting charts! Calculations are the second step (at best).

Additionally there is a negative R2 for Form.A, zero order which "seems" quite odd...

Based on the table you sent I can't conclude that Korsmeyer-Peppas is "better" than Higushi just because the average R2 is larger that the other, without looking/analysing the raw data.

Taking the risk of being unfair (because I haven't see the data) to the people who questioned you about the

It is better to use AIC than R squared for curve fitting in that case you have two models with high correlation R2 more than 0.99