This chapter discusses the factors affecting the performance of machine learning models. The chapter provides technical definitions of performance for different types of machine learning problems. In an e-commerce application, for example, good performance might mean returning correct search results or presenting ads that site visitors frequently click. In a genetic problem, it might mean isolating a few genes responsible for a heritable condition. The chapter describes relevant performance measures for these different problems.
The goal of selecting and fitting a predictive algorithm is to achieve the best possible performance. Achieving performance goals involves three factors: complexity of the problem, complexity of the algorithmic model employed, and the amount and richness of the data available. The chapter includes some visual examples that demonstrate the relationship between problem and model complexity and then provides technical guidelines for use in design and development.
The algorithms covered in this book address a specific class of predictive problem. The problem statement for these problems has two types of variables: