In our research, four variables including pH, initial concentration, contact time, and adsorbent dosage were input and pollutant removal efficiency was output. We have built the ANN model with MSE = 4.21 and R2=0.998. However, we still do not know how to determine the importance of each variable to the output from this model. Could you give us some advice to solve this problem?
Dear Vinh Dinh Nguyen ,
It is an important question and the answer is simple. You have four variables pH, initial concentration, contact time, and adsorbent dosage. You can simply do the following: either remove each time a variable or keep the variable, but scramble it (mean change values to useless random values) run the model and the check the results. This way you can know which variable has the most influence on the accuracy of the model.
To identify the importance of each feature, you can use the Boruta algorithm or recursive feature elimination method. Both methods are implemented in R packages. Not sure about Matlab.
An example is given in Article Ensemble models for solar power forecasting—a weather classi...
This can help you: https://in.mathworks.com/matlabcentral/answers/299646-how-to-obtain-the-relative-importance-of-each-input-variable-for-a-neural-network
You can use Garson's algorithm (weight method), it is easy to implement under Matlab. You can contact me in private.
Dear Vinh Dinh Nguyen
You can use perturbation strategy to find the most influential factor.
Another way to determine the rank of process parameters is the use of Taguchi analysis (statistical method).
This article may help you
Article Optimization of dark fermentative hydrogen production from o...
For more detail please contact me
Thank you
- Badges
- Science topic