Following in the footsteps of its rivals, Amazon Inc. is launching a new cloud-based machine learning service meant to help automate the process of parsing the vast amounts of data produced on the Web. The target audience are the developers building the applications that make use of that information.

That’s best reflected in the inclusion of three ready-made analysis models that can be run without requiring an understanding of the underlying mathematical concepts. The algorithms cover three of the most basic applications of machine learning, which correspond to the top use cases for the science in the public cloud.

The first and most straightforward option is binary classification, which involves determining the likeliest of two outcomes. That’s handy for automating repetitive checks like determining whether or not a transaction is fraudulent. More complex operations can take advantage of multiclass classification, an expanded model that – as the name suggests – can handle more choices.

The third and final use case that Amazon Machine Learning supports out of the box is regression, a technique for quantifying data. One potential use of regression is estimating how many units of a certain product will sell on a given day based on demand from previous years, which could help retailers manage their inventory accordingly and avoid running out of stock early.

Running the pre-packaged models is a simle matter of uploading the target data to Amazon’s cloud, adding the files to the service and setting the desired accuracy, which developers can express in the form of an adjustable cut-off value for probable success. That relative simplicity sets the service apart from the alternatives offered by Microsoft and Google, which are focused on data scientists who can implement their own algorithms.

But Amazon does provide the option to execute custom math in to accommodate customers with more specialized needs than its built-in algorithms can handle. Seeing that the emphasis is on developers, there’s a good chance that the company will go down a similar path  as Microsoft. and provide a marketplace where users can buy pre-packaged models from third parties.

Amazon is asking 10 cents for every 1,000 real-time predictions, plus an hourly charge of one-tenth of a cent per 10 megabytes of memory allocated to executing models. There’s also a batch option that allows for delayed bulk runs without the extra fee, although users still have to pay 42 cents for every hour spent on developing and improving algorithms on the service either way.

Matt Wood – AWS Summit 2015 – theCUBE

Matt Wood, General Manager of Data Science from AWS, talks about the new announcement on @theCUBE live broadcast at #AWSSummit in San Francisco.

Image via Pixabay/h6>