If we are lucky enough to possess sufficient domain knowledge, we can apply it in creating domain-specific features; we utilize our business experience and insights to identify what is in the data and to formulate what from the data correlates to the prediction target. For example, in Chapter 9, Stock Price Prediction with Regression Algorithms, we designed and constructed feature sets for the prediction of stock prices based on factors that investors usually look at when making investment decisions.

While particular domain knowledge is required, sometimes we can still apply some general tips in this category. For example, in fields related to customer analytics, such as market and advertising, the time of the day, day of the week, and month are usually important signals. Given a data point with the value 2017/02/05 in the date column and 14:34:21 in the time column, we can create new features including afternoon, Sunday, and February. In retail, information over a period of time is usually aggregated to provide better insights. The number of times a customer visits a store for the past three months, or the average number of products purchased weekly for the previous year, for instance, can be good predictive indicators for customer behavior prediction.