Azure ML is part of Cortana Intelligence, a series of tools to help organisations get insight out of their data.
Why use Azure Machine Learning?
- Azure ML supports 25 main algorithms already used and proven in Microsoft products like Xbox and Bing search. It's reliable.
- Azure ML provides a facility to get data from the cloud, work on the cloud and get the result out to the cloud. This means teams can share a workspace and collaborate with each other from everywhere.
- Azure ML supports the two main machine learning languages: R and Python. If you know how to write R or Python, it makes it easy to bring your code into the cloud.
- The output of Azure ML can be viewed in many familiar tools, including Power BI, Excel, web browser, PowerPoint, mobile applications and more. If you’re interested in this area, keep an eye out for future posts, where I’ll get into the detail of connecting Azure ML to other tools (see diagram below).
How Azure Machine Learning can be helpful
Azure ML supports three different types of analysis: descriptive analysis, predictive analysis and prescriptive analysis.
While data is everywhere, actionable intelligence from data is harder to find. Our customers want to use their data to drive reinvention of business processes. To do this it needs to be easier to analyse data – to get from “what happened” and “why it happened” to prediction: “what will happen” and ultimately, know “how to make it happen”.
As the chart below shows, predictive and prescriptive analysis are more difficult, but they provide greater future value.
Descriptive analysis (What’s happening? Or, what happened?)
Descriptive analysis allows you to see the current trend, behaviour and structure of data – you analyse data in order to summarise what happened. The majority of business analytics - such as sales reports, web metrics, and social networks analysis - are descriptive .
One of the main approaches to descriptive analysis is to classify the data into different clusters using the K-Means clustering algorithm (Figure 4).
Clustering algorithms are mostly used for answering questions such as:
- Which other customers have similar preferences to this one?
- What is a natural way to divide this set of customers into groups?
- Which viewers like the same kind of movies?
- What are the most common patterns in petrol price changes?
- What is a natural way to break this set of companies up into groups?
Predictive Analysis (What will happen?)
Predictive analytics is all about understanding the future. Predictive analytics provides actionable insights based on data, through estimates about the likelihood of a future outcome, based on probabilities. Although of course it is important to remember that no statistical algorithm can “predict” the future with 100% certainty.
Other posts on our tech blog about predictive analysis:
- Data science: more than just analysis (an overview of data science, predictive and prescriptive analytics)
- How to train your inbox using Microsoft Azure Machine Learning (part one, two and three)
- Cortana Analytics Suite: first look
Predictive analysis has been used for marketing, finance, maintenance, operational efficiency, energy forecasting, text and speech processing, image processing and computer vision, and so forth. Azure ML provides predefined algorithms that help us to make predictions more easily.
The following types of questions can be answered using predictive algorithms:
- When will this part fail?
- Will my customer leave me for a competitor?
- Will this customer renew their subscription?
- Should this customer receive a promotional offer?
- How likely is this person to make a purchase?
- Does this applicant pose an acceptable credit risk?
- How many bikes will be rented in the next hour?
- What will the consumer demand be in this region over the next month?
Predictive analytics: Classifications and Regression Algorithms in Azure ML
Prescriptive Analysis (How we can make it happen?)
Prescriptive analysis is the best way to see how to make a sale or encourage a customer in the future. Recommendation systems are the another name for prescriptive analysis. Customer activity is used to recommend items and improve conversion in the digital store. The history of previous purchases and interests are used to recommend new products. To make the recommendation we employ both descriptive and predictive analysis several times.
One of the famous recommendation systems is Amazon. Once a person purchases something from Amazon, Amazon recommends other products based on the similarity of the items purchased and the similarity of the customer profile to other customers. We name the first approach Content-Based Filtering and the later one Collaborative Filtering.
Azure ML proposes some APIs that can be customised to support recommendation (prescriptive analysis). Microsoft Azure ML’s recommendations include Item to Item recommendations - a customer who bought this also bought that - and Customer to Item recommendations: a customer like you also bought that.
Existing Azure ML modules can be used to do these experiments (see figure below). In future posts we will also explain how to create a recommendation package in Azure ML.
Microsoft Azure ML is a part of Cortana Intelligence that provides a facility for data scientists, business intelligence experts, and software developers to get insight out of data. Azure ML provides some predefined algorithms for users who are not familiar with machine learning algorithms. Also, it provides a facility for data scientists to use their own code in the cloud. Working with Azure ML is easy and users can create web service from their models. Azure ML supports the three main types of analytics: descriptive, predictive, and prescriptive analysis.
In future posts, I’ll explain in more detail how to create descriptive, predictive and prescriptive analytics in Azure ML. I’ll also show how Azure ML can be connected to Power BI, R, and web applications.