Forecasting in AWS Autopilot

For a low/no-code option for forecasting, connected with the SageMaker ecosystem - I'd suggest using SageMaker Canvas. Canvas explicitly supports forecasting, using internal approaches for "quick model", the or Amazon Forecast back-end when users build a full model. If you're not so concerned about close integration with SageMaker, you could also try using the Amazon Forecast service itself.

Time-series forecasting brings special considerations around train/validation data splitting (e.g. for a fair assessment of performance, you need to take the end of your dataset as the validation period, not just randomly select time points throughout). More timeseries-tailored algorithms may also perform better than plain regression approaches (e.g. auto-regressive algos like DeepAR and CNN-QR).

Because it doesn't cater for these points at the moment, Autopilot today doesn't formally support forecasting as a problem type. You technically could try tackling forecasting as a regression problem in Autopilot, for example by engineering lots of features for day-of-week, month-of-year, etc in your data table... But correctly interpreting model metrics would likely be a challenge, and accuracy may not be as good (depending on the nature of your data of course).

Although Amazon SageMaker Autopilot supports time series data type in input dataset, but the problem type options are only Regression, Binary classification, and Multiclass classification 1. In other word, you can use Amazon SageMaker Autopilot to build machine learning models only for regression and classification problems for time series or any sequence data.

Amazon SageMaker Autopilot is an automated machine learning (AutoML) solution to obtain prediction, not forecasting. Amazon SageMaker Canvas and Amazon Forecast service provide better capabilities to address time-series forecast challenges. Amazon SageMaker Canvas gives you the ability to use machine-learning time series forecasts 2, and Forecast service is a fully managed service that uses statistical and machine-learning algorithms to deliver highly accurate time-series forecasts 3 4 5. Please make an informed decision by considering the costs associated with Canvas 6 and Forecast 7. Canvas can be pricey compared with Forecast.