Machine Learning Forecasting: Lag Features, Train/Test Splits & Error Metrics
Forecast daily demand the way ML teams do: lag, rolling and calendar features, a leak-free split by date, naive and seasonal-naive baselines, MAE/RMSE/MAPE, a LINEST regression on lags, walk-forward validation and drop-one feature importance, in 8 graded labs.
The first lesson is free — no account needed.
About this course
A machine learning forecasting course that predicts one fictional bakery's daily croissant demand the way ML teams do, with every step in a spreadsheet, for analysts, data science students and planners. You turn a daily series into a supervised-learning table with Lag 1, Lag 7 and a rolling mean that does not leak, add weekday dummies and a holiday flag, split six weeks at a cut-off date with no peeking into the future, set the bar with naive and seasonal-naive baselines scored by MAE, RMSE and MAPE, fit a regression on the lags with LINEST, predict 14 unseen days and express the gain as forecast skill, run a 12-step walk-forward validation with an expanding window and rank features by drop-one importance with a chart. No coding required. The first lab is free.
Lessons
- Lag & Rolling-Mean Features: Turning a Series into a TableFree
- Calendar Features: Weekday Dummies & a Holiday Flag Pro
- Train/Test Split by Date: No Peeking into the Future Pro
- Baselines First: Naive, Seasonal Naive & MAE, RMSE, MAPE Pro
- Linear Regression on Lags with LINEST Pro
- Predict the Test Set & Compare Errors (with a Chart) Pro
- Walk-Forward Validation: An Expanding Window, Step by Step Pro
- Feature Importance by Drop-One: Which Feature Earns Its Place? (with a Chart) Pro