Data Mining Practice Test

The key idea is modeling the probability of a binary outcome using multiple predictors. Logistic regression is built for this exact purpose: it estimates the probability that the event occurs given the input variables by applying a logistic (sigmoid) function to a linear combination of those predictors. This keeps the predicted values between 0 and 1, so they can be interpreted as probabilities, and the model can be estimated with maximum likelihood. The coefficients tell how each variable shifts the log-odds of the event, making the relationship easy to understand and interpret. While linear regression would predict a continuous value outside the [0,1] range and isn’t appropriate for probabilities, discriminant analysis relies on distributional assumptions that may not hold and aims more at classification with those assumptions, not as flexible probabilistic modeling. K-Nearest Neighbors is a non-parametric method that can classify without a probabilistic form and can be sensitive to scale and data size. Logistic regression, by contrast, balances interpretability, probabilistic output, and applicability across a range of predictor types, making it the best choice for predicting the likelihood of a binary outcome.