Machine Learning, as a method of data analysis that automates analytical model building, is not new. SAS used machine learning techniques to solve business challenges through useful insights for over 40 years.
What is new is the amount of data that is now available. With many using smartphone devices, various digital touchpoints, digital POS and the advent of social media, the volume of data has increased exponentially. Making sense of the data and discovering hidden patterns is making Machine Learning a valuable tool for many.
Machine Learning vs Statistics
Machine Learning has the same roots as statistics: mathematics. “The key difference is that Machine Learning focuses more on automation, derived from computing and predictive power, while statistics focuses on understanding the data itself, such as probabilities of certain behaviors or a hypothesis that relationships exists,” said Wilson Ho (photo right), General Manager – Hong Kong, SAS.
Often, Machine Learning is used for categorizing or cataloging, predicting outcomes based on identified patterns, identifying unknown patterns and detecting anomalous or unexpected behaviors.
For Machine Leaning to be used effectively, a clear business strategy/objective is important. “Here, firms need to define what problems they want to solve, what actions can be taken and defining measurable outcomes. It is a time and data consuming effort,” Ho said. This is why Ho recommended firms to study existing models.
Different models for different problems
Once the problem statement is defined, Machine Learning uses five models to develop the results. They include Supervised Learning, Semi-Supervised Learning, Unsupervised Learning, Reinforcement and Deep Learning.
Take fraud detection for example. Identifying fraud is becoming increasingly sophisticated as fraudsters find new ways to circumvent anti-fraud measures. Machine Learning using the Supervised Learning model to “learn” from various examples of fraud and historic data to quickly determine fraudulent behavior. The same model is used for credit scoring and customer segmentation.
Unsupervised Learning is used when the data does not have any predefined outcomes. The idea here is to identify hidden patterns in data by allowing the machine to do its own correlations. For example, Market Basket Analysis often uses this model to discover cross-sell and up-sell opportunities that are not obvious.
In marketing, Machine Learning is opening new possibilities. Marketers can now group structured and unstructured text to monitor consumer sentiment and respond proactively to complaints. SciSports in the Netherlands uses SAS Viya to analyze images to help football clubs identify the next football stars.
A leading global bank uses Machine Learning and analytics to tease out new patterns of consumer behavior and offer the right services. The Shangri-La Hotel Group uses SAS CI Suite to offer pre-trip, on-trip and post-trip offers to its Gold Circle VIPs.
Major initiatives like Smart City and Smart Government combine various models to develop new possibilities and outcomes. For example, Machine Learning can be used track and detect lifestyle needs of citizen, and improve all aspects of city life. It can also aid governments to understand changing sentiments and behaviors and develop proactive and highly contextualized services.
Targeting the talent shortage
The biggest hurdle to Machine Learning adoption is talent. Here, SAS addresses this with the SAS Academy for Data Science, which aims to make Machine Learning skills accessible as firms become more data-centric and intelligence-driven.
“SAS offers world-class certification in big data, advanced analytics and data science via the SAS Academy for Data Science to help sharpen attendees’ skills and validate their expertise,” Ho said.