Introduction

Machine Learning has garnered a significant share of recent press coverage in both tech and main street media. It is inextricably intertwined with, and central to, discussion and dialogue on topics ranging from big data in general to Facebook’s threat to privacy, Boston Dynamics creepy robotics, and Google’s exploitation of artificial intelligence for good and ill. As such, it is easy to view machine learning as either sinister or magical — neither of which is true. For today’s business leader, an objective and actionable understanding of machine learning is as important as an actionable understanding of finance and financial management.

In this article, we provide an overview of machine learning for business leaders: what it is and how to think about its applicability to your business.

What machine learning is

Machine learning (ML) is a data-driven system development paradigm. ML systems leverage data models, data analysis and feedback to define and refine algorithms to improve model accuracy and system results.

ML systems work by analyzing data to detect patterns or by applying predefined rules to:

Categorize or catalog like objects

Predict likely outcomes or actions based on identified patterns

Identify unknown patterns and relationships

Detect anomalous or unexpected behaviors

Different algorithms learn in different ways. But in general, as new data are provided to the ML system the system "learns" and the algorithm’s performance improves over time.

Problems suited to machine learning

ML, like other software development paradigms, is not one-size-fits-all — some approaches are better suited to particular classes of problems and not suitable for others.

Machine learning is particularly suited to problems where:

Logical rules are unavailable or insufficient to describe the environment — but actionable rules can be intuited

Next actions are varied and the best action depends on conditions that cannot be identified in advance

Understanding why an outcome is suggested is not as important as the accuracy of the outcome

The data is problematic for traditional analytic methods

Now that you know what machine learning is and how to identify problems that lend themselves to ML solutions, let’s explore the steps to define and conduct an ML project.

How to plan and execute a machine learning project

Well executed ML systems follow these recommended steps:

Define Problem

Prepare Data

Evaluate Algorithms

Improve Results

Present Results

These steps, while seemingly generic and common to traditional software system development, require the perspective and attention gained from experience with ML system development.

The best way to approach machine learning system development is to work through an ML project end-to-end and cover the key steps with an experienced guide or team. Every step, from loading data, summarizing data, evaluating algorithms, making initial predictions, refining and presenting results is improved by experience — much like an ML system.

Accordingly, your first project should be viewed as a learning process to understand the mechanics of machine learning, calibrate your expectations and provide a perspective for setting expectations, interpreting and presenting results from dynamic, learning systems. After tackling your first project with the expert assistance you will be prepared to spot and sponsor the next, more consequential machine learning opportunity.