Hi,

Stratified k-fold cross-validation is the most recommended approach. Holdout method is another technique one can consider.

HTH.

Dr. Samer Sarsam

The question needs to be more clear, since it is unclear which technique or what purpose has been achieved using a new model.

- If new model is about classification - use cross validation or split percentage method

- if new model is about clustering - use silhouette index, homogeneity and heterogeneity indexes

- if new model is about association rule analysis - use confidence and lift metrics

There are several other purposes that can be achieved for big data analytics.

Furthermore, simulation (since it is a way to present a computer model works for something/events/data) can be carried out for validation of the new model.

Adding to Mr.Mahoto reply, in regression - Root Mean Squared Error, Relative Squared Error, Mean Absolute error...

If new model is about transformation or cleaning rules in pre-processing. how can i validate the rules

Kiran,

The model had to be built from the training set that gets prepared using, e.g., data transformation technique (there are many other ways; this depends on the nature of the data and the task as well). Once this achieved, the built model needs to be evaluated using, stratified 10-fold cross-validation, for instance.

Cheers,

Dr. Samer Sarsam