Time Series Forecasting Basics
Forecast demand, traffic or sales with honest baselines, proper backtests and useful intervals
A taste of a lesson
My gradient boosting forecast for weekly sales has lower error than ARIMA. Is that enough to choose it?
Not yet. First, did both beat a seasonal naive baseline, using the same weeks from last year? If not, neither adds much value. Second, was the comparison a rolling origin backtest at the horizon you actually need, say four weeks ahead, rather than one split? Third, check that no lag or rolling feature used data after the forecast date. Finally, compare interval coverage, not just point error. What is the error of a seasonal naive forecast over the same backtest periods?
Written by the teacher as an example. In your lesson the tutor answers your own questions, and like any AI it can be wrong.
What you will be able to do
- Decompose a series into trend, seasonality and noise
- Build naive and seasonal naive baselines to beat
- Apply exponential smoothing and lag feature models
- Evaluate forecasts with rolling origin backtests and scale free metrics
- Produce prediction intervals and explain forecast limits
Lesson plan
- 1 Reading a time series Identify trend, seasonality and noise in a series. Start
- 2 Baselines first Build naive and seasonal naive forecasts. Start
- 3 Smoothing and autoregression Understand exponential smoothing and autoregressive ideas. Start
- 4 Machine learning for forecasting Use lag, rolling and calendar features with ML models. Start
- 5 Backtesting properly Evaluate forecasts with rolling origins and fair metrics. Start
- 6 Uncertainty and breaks Produce intervals and handle structural changes. Start
Try asking
About this tutor
An intermediate tutor for analysts and data scientists who need to forecast numbers over time. You will decompose series into trend, seasonality and noise, build naive and seasonal naive baselines, apply exponential smoothing and understand the ideas behind autoregressive models, then use machine learning with lag and calendar features. Lessons stress evaluation: rolling origin backtests, scale free metrics compared with naive forecasts, prediction intervals and the forecast horizon. You also learn to handle holidays, structural breaks and external drivers, and to explain the limits of any forecast honestly.
Reviews
4.3
3 ratingsSample
- Pablo E.Sample
Our fancy model barely beat seasonal naive once we backtested properly. Humbling, and it changed how we report forecasts.
- Oluwaseun B.Sample
Clear on leakage in rolling features. I wanted more on hierarchical forecasting, which was only touched on.
- Keiko M.Sample
MASE was new to me and makes comparisons across products much fairer. The holiday features lesson was useful.
About the teacher
Probability for machine learning, plus forecasting and anomaly detection
9 tutors 361 lessons taught Sample
I teach probability the way machine learning uses it: random variables, likelihood, entropy and simulation. I also teach two applied areas where probability matters every day: time series forecasting and anomaly detection. My work background is in monitoring and forecasting for operational systems, where wrong alarms and missed incidents both have a cost. I teach through small simulations, coin and...
See Kenta's profile and tutorsMore like this
Other tutors on the same or nearby topics.