Description

Introduction

Time Series Forecasting with AWS SageMaker provides a hands-on approach to building accurate and scalable forecasting models using the SageMaker platform. This course focuses on time series data characteristics, feature engineering techniques, and the application of SageMaker built-in algorithms like DeepAR, as well as custom models. You’ll learn to manage temporal datasets, evaluate model performance, and deploy forecasts in production for real-world applications such as demand planning, financial analysis, and resource optimization.

Prerequisites

Participants should have:

• Basic knowledge of machine learning concepts.

• Familiarity with Python programming and data handling (e.g., pandas, NumPy).

• Working understanding of AWS SageMaker basics.

• General awareness of time series data (optional but helpful).

Table of Contents

1. Understanding Time Series Forecasting
    1.1 What is Time Series Forecasting?
    1.2 Use Cases in Different Industries
    1.3 Challenges and Considerations in Time Series Modeling

2. Overview of AWS SageMaker for Time Series
    2.1 SageMaker Built-in Algorithms for Forecasting (DeepAR+, XGBoost)
    2.2 Using Custom Models (ARIMA, Prophet, LSTM)
    2.3 SageMaker Studio and Pipelines for Forecasting Workflows

3. Preparing Time Series Data
    3.1 Formatting Datasets for SageMaker
    3.2 Handling Missing Values and Outliers
    3.3 Creating Features: Lag, Rolling, Seasonal Indicators

4. Using the DeepAR Algorithm in SageMaker
    4.1 How DeepAR Works
    4.2 Preparing JSON and CSV Input Formats
    4.3 Training a DeepAR Model with Estimator API
    4.4 Evaluating Forecast Accuracy

5. Alternative Forecasting Approaches
    5.1 Forecasting with XGBoost
    5.2 Integrating Facebook Prophet and LSTM Models
    5.3 Comparing Results Across Multiple Models

6. Model Evaluation and Interpretation
    6.1 Metrics: RMSE, MAE, MAPE, sMAPE
    6.2 Visualizing Forecasts
    6.3 Cross-Validation Techniques for Time Series

7. Deploying Forecast Models
    7.1 Deploying to Real-time Endpoints
    7.2 Batch Transform for Bulk Forecasting
    7.3 Scheduling Predictions Using Lambda and EventBridge

8. Automation and Pipelines
    8.1 Building an End-to-End Forecast Pipeline in SageMaker
    8.2 Automating Retraining with New Data
    8.3 Logging and Monitoring Forecast Model Performance

9. Real-World Project
    9.1 Forecasting Retail Sales with DeepAR
    9.2 Data Preparation, Model Training, Deployment
    9.3 Evaluation and Business Insights

Time series forecasting with AWS SageMaker empowers businesses to make data-driven predictions at scale. By leveraging SageMaker’s powerful modeling and deployment capabilities, you can build robust forecasting systems that adapt to changing patterns and support smarter decision-making. This course equips you to tackle complex time series problems using both built-in and custom solutions within the AWS ML ecosystem.