1. General principles in building a predictive model
There are typically many reasonable ways to construct a model.
Models may different due to their inferential goals or the way the model is collected.
Key choices include how the input variables should be combined in creating predictors, and which predictors should be included.
In classical regressions these are huge issues, if you include too many predictors in a model, the parameter estimates become so variable as to become useless.

2. General principles in building a predictive model
Here are some useful rules when building a predictive model.
Include all input variables that, for substantive reasons, might be expected to be important in predicting the model.
It isn’t always necessary to include all inputs as separate variables- for example several can be averaged or summed to create a “total score” that can be used as a single predictor.
For inputs that have large effects, consider including their interaction as well.

3. General principles in building a predictive model
Here are four useful strategies for decisions regarding whether to exclude a variable from a prediction model based on the expected sign and statistical significance.
Typically at the 5% level; that is, a coefficient is statistically significant if it is more than 2 standard errors away from zero.

4. General principles in building a predictive model
If the predictor is not statistically significant but has the expected sign, it is generally fine to keep it in. It may not help predictions dramatically but is probably not going to hurt them.
If a predictor is not statistically significant and does not have the expected sign, consider removing it from the model (that is, setting its coefficient equal to zero)
If the predictor is statistically significant and does not have the expected sign, then think hard if it makes sense. Try to gather data on potential lurking variables and include them in the model.
If the predictor is statistically significant and does have the expected sign by all means include it.

5. General principles in building a predictive model
These strategies do not completely solve all the problems but help keep us from making mistakes and discarding important information.
They are predicated on thinking hard about a relationships before fitting the model.
Subject knowledge is always key to helping with the pre-estimation thought process of modelling.
This is why the first stage to any good empirical project is to assess the current subject knowledge, usually with an exhaustive literature review.