Feature construction is a commonly used type of feature engineering, and  people usually refer to it when they talk about feature engineering. This technique is the process of handcrafting or constructing new features from raw data. In this type of feature engineering, domain knowledge is very useful to manually make up other features from existing ones. Like other feature engineering techniques, the purpose of feature construction is to increase the predictivity of your model. A simple example of feature construction is using the date stamp feature to generate two new features, such as AM and PM, which might be useful to distinguish between day and night. We can also transform/convert noisy numerical features into simpler, nominal ones by calculating the mean value of the noisy feature and then determining whether a given row is more than or less than that mean value.