Learning the Ropes: Machine Learning
8 December, 2015
Posted by Jon Rimmer
There is a growing recognition that Machine Learning has a significant role to play in a wide range of applications including data mining, natural language processing, traffic flow optimisation and predictive analytics.
As with any technology trend, machine learning has become somewhat of a buzzword that gets thrown around and often gets confused with the more traditional predictive analytics.
So What Is Machine Learning And Why Is It Different?
Put simply, Machine Learning is a technique that provides an application with the ability to learn and improve without being expressly programmed. Predictive analytics is a use of Machine Learning technology where the application is used to make better decisions about a desired outcome, compared to traditional predictive analytics. Traditional predictive analytics relies on human analysts to test the relationship between all of the available variables and the desired outcome. That analyst has to determine which factors have the strongest influence on the outcome and thus are the best predictors and what combination of those will give the best model.
Machine Learning reverses this technique and starts by measuring the outcome. The program will then automatically uncover which variables are influencing the outcome and what the relationship is between these variables. It can automatically analyse hundreds of possible variable combinations, interactions and responses, with the result giving a model that is substantially more accurate and that can continuously refine and improve itself over time. Machine Learning programs detect patterns in data and adjust the program’s output to deliver the optimal response.
One of the major issues facing machine learning applications is how to interpret non-Boolean data, which is data that can’t be interpreted as a 1 or 0. Examples include describing data where it may be considered cool, cold, warm or hot or anywhere in between. In real life it would be very difficult to define a specific temperature where it goes from cool to cold – it comes down to the opinion of the user. To address this issue, techniques such as Fuzzy Logic can be applied to machine learning applications. These techniques define how the application should handle data that is on a continuous scale and how to handle the boundaries between categories so that there is a natural transition from one to another, e.g. from cool to cold.
Being able to process this type of variable and include it in the machine learning application significantly expands the amount of available data and hence predictor variables, delivering more accurate and reliable predictive models.
It’s easy to see how Machine learning quickly delivers efficiency gains as well as rapid performance improvements over the manual processes involved in traditional predictive analytics. There is an endless list of business applications from improving credit decisioning and debt recovery to determining which display advert to show to a prospective customer at what time.
Logical Glue enables organisations to create and deploy the best predictive models to drive better business performance. To find out more about how Logical Glue can help your business, why not give us a call on 020 8720 7246 or email us at: firstname.lastname@example.org