The output parameter can be estimated using the input variables where each input is dropped one at a time. Alternatively, the output parameter may be approximated regarding just one input variable in each stage. Lastly, the performance of each model is evaluated, in doing so, the significance of each input variable is determined.

The way to tell whether one independent (input) variable Xi has a big or little effect on the dependent (output) variable Y is the use of the so-called elasticity coefficient. The elasticity ratio has the advantage that it is dimensionless (units-free).

If Xi changes by 1% what will be the percentage change in Y?. Elasticity coefficient for Xi is dY/dX*X/Y. Here the derivative dY/dX is the coefficient for the variable X.

I guess your inputs are normalised thus having the same scale. So if you're performing a regression task, each coefficient determines the importance of the feature it's associated with.

For sci-kit models, you can get the importance after the training ("model.fit()'") by:

' importance = model.coef_ ' #Linear or logistic regression

' importance = model.feature_importances_' # for the others

You can use permutation feature importance.

Dear Nandhakumar Murugan

It may be used to choose features by assessing the Information gain of each variable in relation to the target variable.

Kind Regards

Qamar Ul Islam

One option can be used the corrplot Matlab function

https://au.mathworks.com/help/econ/corrplot.html

And also I agree with Hamed's suggestion to check the performance of the model with each input one by one. It would be better to keep the same random seed for all inputs to provide a fair comparison or repeat the process of prediction several times.

Random forest importance

Look for coefficients if you are doing linear regression or logistic regression. Also look for correlation between the variables and check multicollinearity as well.

machine learning declaring the variable is essential

Hi

The systematic way to do this, is to use PCA (principal component analysis). This works if there are some dominant linearity in the data. If nonlinear, there are variations of PCA for nonlinear analysis.

Further, as suggested you can drop inputs, train model and check accuracy of the prediction. You can do this in a more systematic way by comparing metrics like "Information Gain" for each variable.

Dear Nandhakumar,

A possible way of determining the relative importance of the variables might be dropping variables and either computing the Akkaike Information Criteria of each model or performing permutation tests, permuting one input variable at a time B times, producing B measures of model performance, which can then be used to produce an empirical distribution of performances and a heuristic p-value (percentage of models that give better performance than our non-permuted model).

Other ways are to look at coefficients and p-values of coefficients: we look for large coefficients that have significant p-values, as large coefficients with non-significant p-values might or might not be driving the model, while significant p-values with small coefficients only tell us that the effect is significantly different from 0 but that the effect is small.

Best

Dear Nandhakumar Murugan

This link can help you

https://github.com/afai79/AI-and-ML

https://au.mathworks.com/help/econ/corrplot.html

Deat Dr Nandhakumar Murugan . I agree with Dr Anuraj Nayarisseri .

Those columns that are the inputs are referred to as input variables. Whereas the column of data that you may not always have and that you would like to predict for new input data in the future is called the output variable. It is also called the response variable Nandhakumar Murugan

https://machinelearningmastery.com/calculate-feature-importance-with-python/