Model Performance Management with Explainable AI

Chapter 3. The Machine Learning Life Cycle

Now that we have discussed the importance of explainability, let’s dive into the machine learning life cycle to shed some light on what a model goes through on its journey from conception to production. There are various stages in the ML life cycle, and based on the complexity of your domain and/or the maturity of your system, you might already have incorporated many of these stages into your workflow. Depending on the size of your teams and the seniority of the individuals on those teams, each stage may be the responsibility of one team, or the entire life cycle may be managed by one team, or anywhere in between. MPM fits nicely into the model development, deployment, and monitoring stages and can help with monitoring and managing models that are being deployed. We’ll discuss this in more detail in Chapter 4; for now, we’ll focus on the various stages in the ML life cycle and what each one is about. First, though, we will briefly walk through the different types of analytics to illustrate how machine learning has evolved over time.

The Three Types of Analytics

As you can see in Figure 3-1, there are three types of analytics: descriptive analytics, predictive analytics, and prescriptive analytics. Machine learning models can be used for all three.

Descriptive analytics, as the name suggests, usually comes in the form of reports or charts where insights can then be derived by a human. This is the most commonly used of the three types, ...