Key concepts
The typical analyst who has been doing predictive modeling for a while has constructed tens, perhaps hundreds, of linear regression models over the years. If you worked for a large accounting firm in the late 1980s, as I did, and you were doing forecasting, you may have spent your whole day, every day, specifying linear models. You would have run all conceivable permutations of independent variables and transformations of dependent variables, and diligently looked for evidence of heteroscedasticity (non-constant variance in residuals) or multicollinearity (highly correlated features). But most of all, you worked hard to identify key predictor variables and address any bias in your parameter estimates (your coefficients or weights).
Key assumptions of linear regression models
Much of that effort still applies today, though there is now much more emphasis on the accuracy of predictions than on parameter estimates. We worry about overfitting now, in a way that we did...
