A binary (Bernoulli) random variable takes exactly two values, conventionally 1 and 0. In binary classification, 1 denotes the positive class and 0 the negative class. If the probability of 1 is p, the probability of 0 is 1−p. section 3.3.2 section 11.4.2
For an observed binary outcome yᵢ and predicted probability pᵢ, this expression gives the probability assigned to that outcome. section 3.3.2
For independent observations, multiply their individual outcome probabilities to obtain the likelihood of the data. In classification, maximizing this likelihood is equivalent to minimizing binary cross-entropy. section 3.3.2 section 11.4.2




