Amazon Forecast Deprecated2018年〜
A fully managed service that uses machine learning to forecast time series data
What It Does
Amazon Forecast is a machine learning service that predicts future values based on time series data. When you input historical sales data, inventory data, or demand data, it automatically tries multiple machine learning algorithms and generates the most accurate prediction model. No machine learning expertise is required.
Use Cases
It is used for retail demand forecasting, inventory optimization, website traffic prediction, electricity demand forecasting, and workforce planning - any business that needs predictions based on time series data.
Everyday Analogy
Think of it like a weather forecast. It analyzes historical weather data (time series data) to predict tomorrow's weather (future values). Forecast uses sales and demand data instead of weather data to provide a 'weather forecast' for your business.
What Is Forecast?
Amazon Forecast is a machine learning service specialized in time series forecasting. Simply upload CSV data to S3, and it automatically evaluates multiple algorithms (DeepAR+, Prophet, ARIMA, etc.) and selects the optimal model. Forecast results can be retrieved via API and integrated into applications.
Datasets and Forecasting Workflow
Using Forecast involves three steps: importing datasets, creating a predictor, and generating a forecast. In addition to time series data (timestamps and values), you can add related data (weather, promotion information, etc.) to improve accuracy. The AutoML feature automatically selects the optimal algorithm and hyperparameters. For a deeper understanding of datasets and the forecasting workflow, books (Amazon) are also helpful.
Getting Started
Create a dataset group in the Forecast console and import a CSV file from S3. Creating a predictor triggers AutoML to automatically train a model. After training completes, generate a forecast and export results to S3 or retrieve them via API. Note that Amazon Forecast stopped accepting new customers in 2024, and SageMaker Canvas time series forecasting is recommended for new projects.
Things to Watch Out For
- Amazon Forecast stopped accepting new customers in 2024. Consider using SageMaker Canvas for new projects
- Forecast accuracy depends heavily on data quantity and quality. At least several hundred data points of time series data are recommended
