Take a dive into any discussion about predictive analytics, and it is likely that you will find the terms machine learning and analytics interchanged regularly. It is understandable given that both are related, but they are not the same thing. However, Sam Underwood, VP of business strategy with Futurety, a data analytics and marketing agency, points out that in practical terms, the two should work together.
The main difference between machine learning and data analytics depends on which direction you want to look — forward or backward. Data analytics in its simplest form is looking back at what was done to find trends that may help you moving forward. Data analytics is largely focused on identifying key variables that likely caused a result to happen the way it did.
Machine learning, on the other hand, takes all these past variables and applies them to future situations using algorithms to predict future outcomes. The more data we can feed into a machine learning algorithm, the more accurate the algorithm is likely to be.
It's important that these two roles work in tandem and not allow machine learning to replace standard data analysis. “It's always worthwhile to regularly check back on the results of a machine learning process to ensure it was accurate, and if not, consider re-engineering the algorithm for improved performance going forward,” Underwood said.
Related Article: Github's Top Open Datasets For Machine Learning
Machine Learning-Driven Analytics Tool
Historically, data analytics has created a cottage industry of reporting tools that helps companies understand what happened. Predictive analytics takes that a step further, using statistical models to estimate the likelihood of an outcome given a set of input variables to understand what happens next. Michael Murff, head of neuralytics at InsideSales, pointed out that where artificial intelligence (AI), or machine learning, differs is how it uses bigger datasets and more advanced algorithms to make robust predictions of future events. Think of the Waze App, which leverages changing driving conditions to re-route you on the fly.
However, AI and machine learning often come together when both are built into systems that are designed to continuously collect data in order to learn and understand in real-time what should happen. This is a big change from the days of descriptive reports in Excel designed to tell you what happened, and also an improvement on predictive analytics such as a one-off credit scoring decision.“AI rides the wave of faster computing, better algorithms, bigger datasets to build high performance and autonomous systems to solve dynamic problems,” he said.
Murff also pointed out that more and more companies are building AI into decision-management frameworks to optimize their business in ways unimaginable only a few decades prior. The result is the emergence of a whole range of what might be described as machine learning-driven analytics. According to Joe Dumoulin, chief technology and innovation officer at Verint Intelligent Self-Service, this can be broken down into four different models.
- Supervised learning: Think of this as regression analysis on steroids — this form of AI/machine learning can be expensive as it is generally human-intensive. Data is labelled as being in one category or another to formulate a supervised learning algorithm. Then new data is classified according to the model. This is so valuable in business, it is incorporated into things such as Microsoft Excel (formulas).
- Unsupervised learning: Sometimes called factor analysis — differs from supervised learning in that data classifications are not established by a human, and the algorithm employed looks for correlations and patterns to cluster datasets that go with others — i.e. in a retail setting, what items are purchased in conjunction with other items, or for anomaly detection such as fraud, where we are looking for unusual patterns. This is like IBM Watson or Amazon.
- Semi-supervised learning: Used to develop techniques to offer up suggestions/recommendations for labels to a human for datasets. Think of this as a hack to a supervised learning problem — without as much human effort needed.
- Reinforcement learning: Used to train a system to perform a task where there are steps that require evaluation, as each step creates dependencies on the next step, and then manages the interaction of these steps. Examples of applications: self-driving vehicles, robotics, conversational interfaces.
Dumoulin differentiates predictive analytics from AI/machine learning as follows:
- Predictive analytics: Used to formulate a hypothesis based on past experience as indicated by data trends. Answers the question: What does a specific data set/sets tell us?
- AI/Machine learning: Various models as seen above, all involving prep work to condition data and the use of algorithms to generate a model and seek out correlations, patterns and relationships in data. Answers the questions: What are the data sets? What can they tell us?
Predictive Analytics and Machine Learning Are Better Together
Machine learning is essentially a means to predictive analytics as machine learning is used to perform predictive analytics. In fact, most traditional predictive models, like linear regression, can be deployed using machine learning algorithms. Think of predictive analysis as the destination — the outcome we desire — and machine learning as a freeway to get us there, or the means to achieve that outcome. You may still need to use maps, implement quick thinking and take some back roads, but taking the freeway of machine learning will help you get to solid predictive analysis sooner than before.
The last word goes to Doug Bordonaro, ThoughtSpot’s chief data evangelist, who pointed out that the two are close for a number of reasons. He explained that both analytics and machine learning are umbrella terms, giving cover to a number of different topics like predictive analytics, operational analytics, AI and neural networks. “There’s certainly some overlap, as machine learning techniques have a lot of applicability in predictive analytics,” he said.
“I think of analytics as a discipline, bringing order and meaning to data by applying technology to solve business problems. Machine learning offers some tools to help with that, and it's an area that's evolving rapidly.”