What is Machine Learning?
In a previous blog I discussed Big Data, what it is and how it can be used to benefit your business. The main purpose being to extract value from data by gaining insight e.g. “If I discount my product by 5%, I’ll increase sales by 10%”.
The technique used to gain this insight is Analytics and there are several types which I will discuss below:
Descriptive: What happened?
This type of analytics usually involves reports that tell you what happened, for example, comparing sales this week compared to sales last week or the same week last year.
Diagnostic: Why it happened?
With Diagnostic Analytics, typically reports or dashboards are used in conjunction with OLAP cubes which allow users to slice and dice, drill down and investigate the data to be able to identify patterns, trends and correlations etc.
Predictive: What might happen?
Predictive Analytics is a way of using data to enable businesses to identify potential opportunity or risks associated with a set of conditions which enable proactive decision making as opposed to making assumptions. For example, sending marketing campaigns to customers who have the highest probability to buy.
Machine Learning (ML) falls under the Predictive Analytics category
Prescriptive: How can I make it happen?
Prescriptive analytics is a relatively new form of analytics which aims to allow users to choose a number of different possible actions based on predictions and offers recommendations to take advantage of these predictions to guide them to a solution. Where predictive analytics can highlight future trends in sales or marketing, prescriptive can provide the ability to actually achieve them.
Machine Learning in a Nutshell
Machine Learning is a subset Artificial Intelligence (which I will cover in a future blog post) and its goal is to enable computers to learn on their own. Learning algorithms are used to enable a machine to identify patterns in data, build models that explain these patterns and then predict what might happen without having to pre-program rules.
To give this a little context, let’s look at an example of a real-life decision making process. Before we consider buying a product, usually we look at reviews written about that product which describes how good or bad it may be. If most of the reviews have words like “bad quality”, “not good” etc. then we’ll probably look for another product. Alternatively, if they say “great”, “good quality” then perhaps we’ll go ahead and purchase that product. The reviews help us make a decision to act based on the pattern of words used. In this example, the product reviews written by buyers of that product influence future purchases of others.
Essentially, Machine Learning focuses on developing computer programs into algorithms which attempt to mimic this decision-making process, based on data, to make accurate predictions.
How does it work?
Generally, computers are not smart and they require instructions in order to do anything, so there are a few essential requirements in order to get started with Machine Learning:
To begin, there need to be a goal of some kind, a question that needs answering. Then you need data, and a lot of it (here’s where big Big Data comes into play), to accurately predict based on historical events. Then someone with knowledge and expertise to be able to verify the answer is correct. Finally, a pattern. If there is no pattern, perhaps there isn’t enough data or the data quality is poor/incomplete.
Types of Machine Learning
There are three main types of machine learning:
This is when example data is used to learn and is similar to human learning under the supervision of a teacher. The teacher provides good examples for the student to memorize, and the student then derives general rules from these examples.
Typical use cases for this kind of Machine Learning would be where you would like to determine whether an email is genuine or spam, or if a football team will win or lose. Another would be for spotting a trend in stock market prices or the weather forecast.
This type of learning finds patterns based on data input without any examples or instructions on what to do with it. The goal is then to find some structure or similarities in the data. A classic example of unsupervised learning is a recommendation system, for example, after you purchase a product, you’re often suggested by other products you might be interested in.
This form of Marching Learning is basically learning from trial and error. An example of this would be In where a robot picks a product from a box, capturing the image of that object and puts it into a container. If it succeeds or fails it store the image of the product and gain knowledge and train itself to do this job better with more speed and precision.
The critical ingredient to machine learning is having a clear goal and data, together these make a recipe for success in your business. If you’re interested in learning more about Big Data and using Machine Learning in Microsoft Azure contact us today!