Amazon Forecast
Overview
Amazon Forecast uses historical time series data combined with user provided parameter data to generate predictions. The service requires time series data as an input. This can be argumented with local weather data. The desired quantile or mean forcast can be selected and the forcast is output as a CSV file. The output is accompanied by a explainability report for explaining driving factors such as price or weather. The service automatically trains and deploys the model. It also tests the model to track it’s accuracy over time to identify re-training needs using a technique called backtesting. Backtesting involves comparing the models future prediction against the actual outcome.
Amazon Forecast allows forecasting to be performed without having to set up AWS infrastructue an train a model. This is termed as AWS doing the heavy lifting. However if more control over the model is required the Sagemaker built in algorithms can be used, particularly the Deep AR Forcasting algorithm.
- AWS docs: https://aws.amazon.com/forecast
- AWS FAQs: https://aws.amazon.com/forecast/faqs/
Key features of Amazon Forecasting
- Works with any historical time series data to create accurate forecasts
- Automated machine learning
- Based on the same technology used at Amazon.com
- Easily evaluate the accuracy of your forecasting models
- Visualize forecasts
- Integrate with your existing tools
- Generate probabilistic forecasts
Amazon Forecast Use cases
- Product Demand Planning
- Financial planning
- Resource planning
Video: How Taco Bell is improving digital availability with ML forecasting
Credits
- Photo by Isaac Smith on Unsplash