Feature Engineering and Selection

Abstract

Building Machine Learning systems and pipelines take significant effort, which is evident from the knowledge you gained in the previous chapters. In the first chapter, we presented some high-level architecture for building Machine Learning pipelines. The path from data to insights and information is not an easy and direct one. It is tough and also iterative in nature involving data scientists and analysts to reiterate through several steps multiple times to get to the perfect model and derive correct insights. The limitation of Machine Learning algorithms is the fact that they can only understand numerical values as inputs. This is because, at the heart of any algorithm, we usually have multiple mathematical equations, constraints, optimizations and computations. Hence it is almost impossible for us to feed raw data into any algorithm and expect results. This is where features and attributes are extremely helpful in building models on top of our data.