prerequisite for the variables for the ANN model

As in statistics, if we apply regression, we conduct correlations or sensetivity analysis or distribution of data etc.

are there any prerequisites or tests before applying the ANN model. should we include all the explanator variables in the ANN model or any criteria for selection of variables for the model?

[KSSV]

If your variables are x1,x2,x3 and taget is y, in ANN we try to get:

y = f(x1,x2,x3)

You should understand that, x1,x2,x3 should vary y. Is it so? Acorellation study will help for this.

[Hassam Jadoon]

Is there any chance of multicollinearity if all the variables are used which high pearson correlation?

Answer 1

[Ayush Aniket]

Yes, there are important prerequisites before applying an Artificial Neural Network (ANN). Although ANNs can model complex relationships, it's essential to perform exploratory data analysis, check for missing values, outliers, and understand variable distributions. Feature scaling is also recommended to improve training stability.

Not all explanatory variables should be included. Including irrelevant or highly correlated variables can lead to overfitting or multicollinearity. Feature selection methods like correlation analysis, variance inflation factor (VIF), LASSO regression, or principal component analysis (PCA) help identify useful predictors.

Each input variable should ideally influence the target. Correlation studies can help assess this, but keep in mind that ANNs can capture nonlinear relationships that simple correlations may miss.

Multicollinearity can still affect ANN performance by making training unstable or reducing interpretability. If predictors are highly correlated, consider removing or combining them, or applying dimensionality reduction techniques.