Amazon Web Services Inc. (AWS), announced the availability of Amazon Forecast. This tool uses machine learning to refine its prognostication capabilities.
Amazon Forecast does not require users to have machine-learning skills. Users need only input historical data and any associated data, such as variables that could affect predictions. This metadata is combined with historical time-series information and can include store sizes or widget colors. Time-series data can be used to track everything from product sales and inventory levels to Web site visits per hour and more.
Although it is a good fit for retail sales forecasting, AWS claims that its Forecast service can also be used for logistics, finance, and advertising.
When time-series data is combined and that additional metadata, which provides additional variables for machine-learning algorithms to crunch, business predictions can be up to 50% more accurate.
The machine learning function gives you an advantage over other tools that forecast business trends. These tools can be anything from simple spreadsheets to enterprise software.
[Click on the image to see a larger view.] Amazon Forecast (source AWS) “These tools build predictions by looking at historical series of data, which are called time series data,” AWS stated. These tools might attempt to predict the future sales for a raincoat by only looking at the sales data. The assumption is that the past determines the future. This approach is not able to accurately forecast large amounts of data with irregular trends. It is difficult to combine data that changes over time (such price, discounts, and number of workers) with relevant independent variables such as product features and store location.
The fully managed service is now available in the Northern Virginia and Ohio, Oregon, Tokyo as well as Singapore and Ireland AWS regions.