Learn end-to-end time series forecasting, from preparing data and identifying trends, seasonality, and stationarity to building forecasts with spreadsheets, BI tools, Python, classical statistical models, neural networks, and foundation models. Evaluate forecast accuracy using appropriate validation strategies, error metrics, diagnostics, and prediction intervals while communicating uncertainty to stakeholders. Compare forecasting approaches based on interpretability, scalability, and business requirements, then apply these techniques in a capstone project to recommend a forecasting strategy for real-world decision-making.
Overview
Syllabus
- Course Intro
- Learn time series forecasting, from spreadsheets to AI models, to make better business decisions by evaluating forecasts, quantifying uncertainty, and communicating insights.
- Introduction to Time-Series Data
- Identify trends, seasonality, and noise in time series to forecast future values while accounting for uncertainty and the differing costs of over- and underforecasting.
- Forecasting in Spreadsheets
- Apply naive, moving average, and trend forecasts in spreadsheets, compare accuracy with holdout validation, and recognize the limits of simple forecasting methods.
- Understanding Forecast Uncertainty
- Understand prediction intervals to communicate forecast uncertainty, explain why intervals widen over time, and make better planning decisions based on forecasting risk.
- Forecasting with BI Tools
- Generate automated forecasts with BI tools, interpret ETS-based predictions, choose the right data granularity, and recognize the limits of black-box forecasting.
- Classical Statistical Models: When Spreadsheets Aren't Enough
- Understand the differences between spreadsheet, BI, and classical forecasting models, including their strengths, limitations, and appropriate use cases.
- Understanding Time-Series Data
- Understand how time series frequency, granularity, and time indexing affect forecasting accuracy, seasonality, and model reliability.
- Representing Time-Series Data in pandas
- Prepare time series data in pandas by creating a DatetimeIndex, resampling data, validating quality, and engineering features for forecasting models.
- Rolling Behavior and Stationarity
- Evaluate stationarity using rolling statistics and the ADF test, then apply transformations to prepare time series data for reliable forecasting models.
- Transforming Time-Series Data in pandas
- Transform time series data by testing for stationarity and applying differencing or log transformations to prepare data for forecasting.
- Time-Series Model Components
- Understand ARIMA and SARIMAX model components, select model orders with ACF and PACF, and incorporate seasonal patterns and external variables.
- Time-Series Modeling with statsmodels
- Build ARIMA and SARIMAX models in statsmodels, compare models with AIC and BIC, and generate forecasts with prediction intervals.
- Interpreting Forecasts and Model Assumptions
- Evaluate forecast reliability using residual diagnostics, prediction intervals, and error metrics to determine whether a model is trustworthy for business decisions.
- Running Time-Series Diagnostics in Python
- Validate time series models using residual diagnostics, AIC/BIC, and forecast error metrics to select reliable models for production forecasting.
- Prophet as an Auto-Tuned Classical Model
- Use Prophet to generate scalable forecasts, incorporate seasonality and external regressors, and compare its strengths and trade-offs with ARIMA and SARIMAX.
- Introduction to Deep Learning for Time-Series Forecasting
- Understand when to use deep learning for time series forecasting and compare its strengths, limitations, and trade-offs with classical forecasting models.
- Training a Neural Forecasting Model
- Train neural forecasting models with N-BEATS, generate prediction intervals using quantile regression, and compare neural forecasts with classical models.
- Foundation Models for Time-Series Forecasting
- Apply foundation models for zero-shot forecasting, recognize when they outperform specialized models, and evaluate their trade-offs in accuracy, transparency, and control.
- Zero-Shot Forecasting with a Foundation Model
- Apply zero-shot forecasting with foundation models, evaluate forecasts through backtesting, and compare results using forecast error metrics.
- Course Review
- Review the complete time series forecasting workflow, from data preparation and model selection to uncertainty, diagnostics, and business decision-making.
- Time Series Forecasting Project
- Apply time series forecasting techniques to evaluate multiple models, communicate results, and recommend a forecasting strategy for executive decision-making.
Taught by
Christopher Agostino
Related Courses
Related Articles
-
7 Best Mathematics for Machine Learning Courses in 2026
-
Udacity Just Launched a New Master’s Degree in AI: Here’s a Look Inside
-
Udacity’s Agentic AI Nanodegree Program: A Developer’s Review
-
12 Best Deep Learning Courses for 2026
-
From Zero to GenAI: 9 Unique Ways to Understand Large Language Models
-
7 Best AI Video Generation Courses (Free & Paid)
