Amazon Machine Learning is an Amazon Web Services product that allows a developer to discover patterns in end-user data through algorithms, construct mathematical models based on these patterns and then create and implement predictive applications.
The service helps companies improve the profitability and effectiveness of their applications. For example, models can be used to detect fraudulent charges with online payments, predict items that will interest a particular end user or forecast product demand during a particular timeframe.
A developer sets up machine learning models for applications in accordance with specified needs, eliminating the need for the developer to write custom prediction code or manage the infrastructure. Amazon generates models by using what it calls an "industry-standard logistic regression algorithm," which determines the probability of how an end user will interact with an application based on past data.
A developer can retrieve predictions using the batch API -- for bulk requests -- or a real-time API -- for individual records. The service processes both types of API requests immediately, and can handle up to five simultaneous batch requests.
Amazon Machine Learning reads data through Amazon Simple Storage Service (S3), Redshift and Relational Database Service, and then visualizes the data through the AWS Management Console and the Amazon Machine Learning API. Data from other AWS products can also be exported into CSV files, which can be placed into Amazon S3 buckets to be accessed by Amazon Machine Learning.
A developer cannot import models into or export models out of Amazon Machine Learning.
Amazon Machine Learning models and other system artifacts are encrypted both in transit and at rest. Requests to the service are made using a secure sockets layer (SSL) connection. A developer can also implement Amazon Identity and Access Management policies to further secure applications.