For a linear model to offer reliable predictions, predictors must satisfy a certain number of conditions. These conditions are known as the Assumptions of Multiple Linear Regression (http://www.statisticssolutions.com/assumptions-of-multiple-linear-regression/):
- Linear relationship: The predictors should have some level of linear relationship with the outcome
- Multivariate normality: The predictors should follow a Gaussian distribution
- No or little multicollinearity: The predictors should not be correlated to one another
- Homoscedasticity: The variance of each predictor should remain more or less constant across the whole range of values
Of course, these assumptions are seldom verified. But there are ...