Machine Learning Is Predictive Analytics

The four levels of analytics

To understand why machine learning is a fundamental data science skill, let’s take a step back and consider how data scientists bring value to their organizations.

In a word, data scientists bring value through analytics. Analytics is the systematic computational analysis of data or statistics. It’s used for the discovery, interpretation, and communication of meaningful patterns in data. It also entails applying data patterns toward effective decision-making.

To put machine learning in an analytics context, the following framework for classifying analytics is handy:

• Descriptive Analytics: Using data in traditional reporting and business intelligence (e.g., executive scorecard) scenarios. The analyses conducted are backward looking and provide information as to what happened. For example, reporting quarterly sales over a two-year period.

• Diagnostic Analytics: Using data to go beyond reporting to understand why something happened. Historical data is analyzed with the goal of finding underlying patterns and associations that explain what has been observed.

• Predictive Analytics: Using data to understand what is likely to happen in the future. Trends, patterns, and associations in historical data are discovered using automated processes (e.g., machine learning). The goal is to be able to make accurate predictions to gain insights and streamline processes.

• Prescriptive Analytics: Using data to understand how to optimize processes. Many processes can be represented as systems of inputs, outputs, and constraints. The goal is to create accurate system representations and use data to find optimal configurations of the inputs and outputs given the constraints.

Given the above framework, machine learning is a form of predictive analytics. However, despite media’s focus on machine learning as predictive analytics, it’s far more than that.

Machine learning is more than prediction

Data scientists wanting to have the most impact within their organizations know their work encompasses more than predictive analytics. Yes, predictive analytics is tremendously valuable to organizations. Much press has been given to the success of machine learning solutions in areas like health care and fraud detection. However, machine learning can also be a potent tool for diagnostic analytics.

Take an example from the domain of human resources (HR) of being able to predict that good employees are going to resign from the organization. Machine learning can be used to build predictive analytics for this problem, as we’ll learn later in the course.

Assuming the predictions are accurate, there’s no doubt the solution’s predictions would be valuable to the HR organization. For example, the HR organization could use the predictions to proactively reach out to employees to keep them from resigning.

What is likely more valuable than the predictions themselves, however, is the meaningful patterns in the data that machine learning found (i.e., the why of good employees resigning).

The HR organization could use these patterns to guide new policies, benefits, compensation plans, programs, etc., that could be used to combat the reasons employees resign across the whole organization.

Now that’s value!