Question

What is a sensible approach to building and validating a predictive model

Answer 1

Hi ,
I am sharing some important steps to create and validate a predictive model.

1. Problem definition and data inspection-
First of all , We should be clarify on the problem. And we need to inspect the data like all columns are need to be take care . If some columns are not useful for model. We should remove that columns. ALso we need to check that all the format of each columns are good to go.

2. Coding of predictors

Continuous predictor variables can be coded in many ways. Continuous predictors variable can often be modelled as a linear association in a predictive regression model.
we should search and categorize some predictors for a user-friendly format of the prediction model.

3. Model specification

There are many methods but Selection methods are mostly and widely used to reduce a set of candidate predictors.
When events numbers are low, the selection will be instable, the performance of the model is over-estimated.

4. Model estimation

Regression coefficients should need to be estimated once a model is specified. We mostly estimate coefficients with maximum ML methods For logistic and Cox regression models. The aim to limit overfitting of a model to the available data.

5. Model performance

Mostly researchers may need to specify the quality. Many of the performance measures are mostly and wildly used, like model calibration and discrimination

6. Model validity

There are two types of validity in prective model , 1. Internal validaty and 2. External validaty .Most important to separate them . Internal validity goes for the specified population that the data originated from reproducibility. Mostly we use random sample for model development but suboptimal form of internal validation

External validity may be evaluated by studying . Example patients who were more recently treated or treated in fully different settings Its called strong external validation.