Model Evaluation

Many people try to develop models to perform a certain task (for example, predicting house prices). Oftentimes, these models cannot represent 100 percent of reality. In the real estate example, we cannot exactly predict a house price all the time. However, it does not mean that the model is garbage. In general, all statistical and machine learning models face this problem. Then, why build one in the first place? Even though we cannot represent reality 100 percent, we can still model useful behavior and represent reality closely enough. For example, we can use demographic information like ZIP code to predict price, and this model can perform better than randomness. This fact lays the foundation for this entire chapter. You need to ensure that this idea is planted well because everybody cares about the insight you bring to the table based on your model.