It is a function to measure how bad a classification or regression model is. For Linear Regression problems, it measures the distance between the linear model’s predictions and training examples. For classification problems, it measures how well a set of estimated class probabilities matches the target classes.