Amazon Forecast melds time series data with machine learning
Consider Amazon Forecast to help with data predictions, resource planning and more. Find out how the service works and ways it can be customized to meet enterprise needs.
There are various methods to generate accurate business predictions, but one increasingly stands out: time series forecasting.
time series forecasting, which is often used in astronomy, mathematical finance and other fields, differs from many predictive analysis models because it includes time as a dimension. This means there is an order dependence between observations made from a data set, which makes time series forecasting more accurate when it estimates trends -- especially within broader time periods.
In November 2018, Amazon added Amazon Forecast, a managed time series forecasting service that uses AWS machine learning technology.
How Amazon Forecast works
Because Amazon Forecast is a managed service, users don't need to provision any infrastructure to use it. It's also geared toward those with little or no prior experience with forecasting or machine learning. For example, the service can automatically build and train predictive models, which otherwise requires a lot of experience and time to do.
To use Amazon Forecast, provide time series data -- which can be in one or more data sets -- and then choose an algorithm or let the service choose one for you. Amazon Forecast will take this data and produce the forecasting model. It also provides the expected accuracy of the model, so users can add more data and retrain if needed.
The service also supports custom algorithms, so organizations can import their own from Amazon SageMaker. Another customization option is to take an algorithm from Amazon Forecast and modify it with Amazon SageMaker.
Amazon Forecast use cases and integrations
Amazon Forecast has a number of different uses. For example, it can plan business financials, based on forecasts around crucial information, like expenses, revenue and cash flow. The service is also beneficial for resource planning. Cost control is crucial, and an accurate estimation of a business's needs can save a lot of money. Even if variables here change frequently, organizations can retrain the model on a regularly scheduled basis to keep predictions up to date.
Amazon Forecast will only be as good as the data you migrate into the service. The more quality data you migrate, the better Amazon Forecast can predict IT capacity, logistics, web traffic, manufacturing, travel demand and more. For example, a manufacturer can migrate time series data sets to the service, such as detailed order history. Forecasts will then be able to predict the next batch of order volumes based off that data.
Users should look to integrate Forecast with systems that automatically migrate data into the service, such as Amazon Timestream, a managed time series database service. This integration could help that same manufacturer predict demand for inventory parts and determine how many of each particular part it will need to order.
Amazon Forecast also works with various enterprise applications, such as SAP or Oracle Supply Chain Management.
Like any other AWS service, each AWS account is limited to a specific amount of resources that can be used to run a forecast. For example, the target S3 bucket for training data can hold a maximum 10 GB file.
Amazon Forecast pricing availability
Amazon Forecast follows a pay-as-you go model. Users incur costs for data storage, generated forecasts and any training for custom models. The service is also part of the free tier, which means users can perform some initial tests without additional costs.
The service is, as of publication, available in preview in U.S. East (Northern Virginia) and U.S. West (Oregon), but general availability looks likely for 2019.
