Written by Alan Estes
Machine learning (ML) is taking the IT world by storm. One of the hottest phrases in the 21st century, the term was actually coined in 1959 by Arthur Samuel, who defined ML as a “field of study that gives computers the ability to learn without being explicitly programmed.” While the concept of ML first came to prominence in the 1950s and 1960s, particularly in academia, there were only a few applications that could take advantage of such sophisticated technology at the time.
ML saw a resurgence shortly after the year 2000. With the costs of storage and compute power starting to drop, ML suddenly became an affordable and scalable solution. Today, ML is on the rise, thanks to a few key factors:
- Access to cloud computing
- Open source tools that evolve with technology
- Increased demand for enhanced product development, customer management and process automation
ML has advanced greatly since its origin, and every business venturing into ML needs to understand how to build successful models. As such, it is essential to know what the possibilities of ML are today. Even more important is to understand the common myths associated with ML and how to avoid them in order to best utilize and benefit from its capabilities.
Machine Learning Today
Today, ML is complemented by a plethora of open source workbenches and utilities, including Python, R, TensorFlow, scikit-learn and many more. It’s also bolstered by the current state of cloud computing as well as scalable data and data storage capacities. In frameworks like Keras for TensorFlow, algorithms are contained in prepackaged software libraries and are widely available to anyone who wants to delve into the world of ML.
ML first took hold in academia and it remains a staple of IT learning in colleges and institutions around the globe. With undergraduate and masters programs dedicated to developing skills in data science and ML the technology will only improve in the coming years.
In its current state, ML has many different uses in a multitude of professional disciplines:
- Financial services: ML is useful in credit underwriting, fraud detection, and portfolio risk management.
- Marketing and advertising: Tech-savvy marketers use ML for response modeling, dynamic AdWord bidding and banner ad targeting.
- Healthcare: The healthcare industry relies on ML for drug research and development, pathology screening using computer vision, and genomic/transcriptomic research.
- Retail: ML is used in retail for product pricing, purchasing and logistics.
With all of ML’s potential uses, however, it is important to acknowledge some potential pitfalls so that you know to avoid them in the future.
Myth #1: We have all this data, now all we need is a data science team to make sense out of it and put it all to work.
It’s often thought that a data science team is all that’s needed to make the most of ML. The reality is the use cases must originate from the product or strategy teams. A data science team certainly should be used to push the organization’s agenda forward and help refine the use cases. For best results, connect the data science team with strategy and execution teams to create and establish value.
Myth #2: Machine learning functionality requires expensive infrastructure and support systems.
While there was a time when ML was cost prohibitive, today’s systems are actually quite affordable. Just be sure to establish objectives that align with your resource capabilities. With a reasonable investment, you can accommodate your high-value use cases, which typically consume relatively low volumes of data anyway. With such an approach, you can realize high project ROI.
Myth #3: Machine learning results are “black box” solutions.
Although it is true that MLalgorithms can be quite complex, they are all pretty standard in the worlds of data science and information technology. The rise of explainable AI and relevant literature has given immense insight about what is going on under the hood of widely accepted ML algorithms.
Myth #4: Machine learning applications get smarter and smarter over time.
It’s a common misconception that all ML systems actually gain intelligence over the course of time. While this is possible, certain cases can be architected as “self-learning” processes; however, it’s not the primary goal of every application. The vast majority of ML applications use a static algorithm to classify/predict/forecast with the expectation that model refits or updates will be a manual process.
The complexity associated with self-learning solutions often leads to investments that render the process unprofitable. In short, avoid the allure to develop and deploy a solution that unnecessarily detracts from the business strategy. Static algorithms tend to solve a substantial number of problems.
Myth #5: The data science team alone will develop a self-sustaining machine learning solution.
In most cases, the average data science team will not devise, develop, and implement the end-to-end solution. Not only is a highly qualified data science team required to develop and produce a working model, but product teams are also needed to define business problems and envision solutions. Furthermore, software teams are needed to implement the solution. As such, a complex, multi-stage process, multiple teams within your organization must collaborate consistently to drive results.
The Future of Machine Learning
Modern ML is full of promise. It’s a reliable engine that drives innovation, creates new products, enhances the customer experience and automates tedious tasks, but pioneers of ML need to be wary of the common pitfalls. For more information, or to find out how you can benefit from a team of experts, please contact DecisivEdge today.
About the Author:
For over three decades, Alan Estes has been leveraging data to solve business problems, improve customer experiences, and automate complex tasks. A leader in the Data Science domain, he has successfully developed and deployed numerous machine learning solutions in an open‑source analytical tech stack. Alan has lead enterprise projects in a variety of business functions across the financial services product life-cycle.
Alan is responsible for setting the agenda and driving the sustainable growth of the Data Science practice, ensuring that the company’s data science professionals are devising efficient solutions for our clients and deepening client partner relationships.
Alan lead the Data Science team for the Small Business Banking business at Capital One. He has also worked at Sallie Mae, K2 Financial (a consumer lending start‑up), Bank of New York, FirstUSA/Chase and has served as an economist on the Board of Governors of the Federal Reserve System.
Alan graduated from the University of Redlands with a BS in Economics and holds an MA in Economics from Virginia Tech.