Logistic Regression
The main advantage is that the output of the prediction equation is a probability on a proper 0-1 scale, which means that the authors have some direct and natural measure of the uncertainty in the assignment.
Abstract
The main advantage is that the output of the prediction equation is a probability on a proper 0-1 scale. This means that we have some direct and natural measure of the uncertainty in the assignment. The probabilities should be treated with a little caution, because the logistic model is likely to be an approximation at best, but still they can be very useful. They are certainly better than the not-uncommon alternative approach of using multiple regression and then fudging the predictions on to a 0-1 scale and treating them as probabilities. dent on x" . .., xp through the model above) of resulting in a 1. A fancier name for this is a Bernoulli trial with success probability p. Then we can use the method of maximum likelihood to estimate the coefficients Po, PI, . .., Pp by the values that make the observed pattern of Os and 1s in the training set most probable. Many statistical packages include a routine for logistic regression: it is a standard tool for medical statisticians in problems where 0/ 1 codes for alive/dead. 5 0.0 1.0