The pitfalls of ML

Consider the problem of predicting the weather forecast. We will begin constructing our predictive model by doing some feature selection. With some domain knowledge, we initially identify the feature air pressure as a relevant predictor. We will record different Pa values (Pascals, a measure of air pressure) over different days on the island of Hawaii, where we live. Some of these days turn out to be sunny, and others rainy.