Module 1Forecasts serve decisions
3 lessonsForecasts serve decisions. Meet a summer with too many parasols and too few help-desk agents, frame a forecast by its level, horizon and cost, and open the Larkfield data and the backtest table.
Predict demand, tickets and staff needs well enough to act on them. Compare simple, statistical, machine learning and foundation-model forecasts in one honest backtest, turn the forecast range into a stock order and a staff roster, and run it as a weekly job.
Read every lesson as a guest, or sign in to enroll and track your progress.
Enroll to save the lessons that you complete. It is free.
You are enrolled. From now on, the lessons that you complete are saved.
This course teaches forecasting as a decision tool: reading a time series, simple baselines and honest backtests, statistical, machine learning and foundation-model forecasters compared on the same data, hierarchies, slow sellers and new products, forecast ranges, the stock order and staff roster they feed, and a forecast that runs every week. It does not teach supervised learning from the start, anomaly detectors or the platform that schedules and monitors jobs.
Take another course if you want to:
By the end of this course, you will be able to:
Larkfield ordered 2,000 parasols for a summer that turned out cold, and hundreds of them were sold at a discount in the autumn. In the one hot week of that summer, the help desk had too few agents, and customers waited a day longer for a reply. Both mistakes came from a forecast, and both had a cost.
In this course, you follow Priya, Larkfield’s data scientist, as she builds the forecasts that Kenji uses to order stock and Grace uses to plan the help-desk roster. You start with the simplest forecasts, such as “the same as last week”, and you test every method the same honest way: on dates the method did not see. Each module adds rows to one backtest table, so you can see a simple method beat a complex one on some series and lose on others, and learn why.
Then you turn the forecast into a decision. A forecast is a range, not one number. You choose how much stock to order and how many agents to schedule from that range and from the cost of a lost sale, a left-over item and an unstaffed hour. You finish with a small job that runs every week, watches its own error and records when a person changes the forecast.
The Larkfield data is synthetic: we made it for this course, with the problems that real data has, and we kept its last weeks hidden so that you can score your forecasts as reality would. Every notebook runs on a laptop CPU or on free Colab in a few minutes.
You completed Machine Learning, From Problem to Reliable Model, or you have the same skills. You can read and change Python and pandas code in a notebook, and you know train and test sets, regression metrics and gradient boosting. The mathematics course helps but is not required. You need a free Google or Kaggle account for the notebooks, or Python 3.12 on your computer. You need no key and no money.
Courses to take first: Machine Learning: From Problem to Reliable Model.
Data scientists, ML engineers and analysts who must predict demand, sales, load, tickets or staffing, and the engineers who run those forecasts every week.
Experienced AI practitioners designed this course. AI helped us draft the content; people chose the teaching approach and the examples, and reviewed every lesson. The outputs that you see are the actual results of running the code. How we make our courses
Forecasts serve decisions. Meet a summer with too many parasols and too few help-desk agents, frame a forecast by its level, horizon and cost, and open the Larkfield data and the backtest table.
Read a time series. Split a series into trend, seasons and the rest, measure its memory with lags and autocorrelation, put holidays and promotions in a calendar, and clean missing days and stock-outs without hiding demand.
Baselines and honest testing. Build the naive forecasts, test them with a rolling-origin backtest, and measure the error with MASE, WAPE and bias instead of a measure that breaks on zero days.
Statistical forecasters. Exponential smoothing, ARIMA and SARIMA, and MSTL and Theta for hourly tickets with a daily and a weekly season, each added to the backtest table.
Machine learning forecasters. Forecasting as regression with lag, calendar and promotion features, one LightGBM model for all products, a pretrained foundation model on a CPU, and when a simple method is the better choice.
Hard series and many series. Reconcile product, category and region forecasts so that they add up, forecast slow sellers with Croston, new products from similar ones, and series after a price rise.
Uncertainty and the decision. Turn a forecast into a range, score it with coverage and pinball loss, and choose the stock order and the staff roster from the cost of a miss.
Run a forecast in production. A weekly forecasting job that checks its data and keeps every forecast, a bias alarm, and planners' overrides measured by the value they add.
Three end-to-end case studies on real open data. Bike-share demand, the electricity load of many customers, and an online retailer's sales with a hierarchy. Complete at least one for the certificate; each one is a portfolio piece.
Complete at least one case study for the certificate. The others are optional. What each one covers
Use every module in one practice project. You collect evidence, and check it against the course outcomes.
Realistic projects from start to finish, on real data, with every output explained. Each one makes a portfolio piece. Complete at least one for the certificate; the others are optional.
Bike-share demand and fleet planning2 hr 30 min
Forecast hourly or daily bike rentals with the weather, compare seasonal-naive, MSTL, LightGBM and Chronos-Bolt in a backtest, and decide how many bikes to move.
Electricity load for many customers2 hr 30 min
Forecast many customers' electricity load, compare global and per-series models with quantile forecasts, and price buying too little or too much power.
Retail demand with a hierarchy2 hr 30 min
Forecast an online retailer's sales by country and product, reconcile the hierarchy, handle slow sellers, and set order quantities from a cost table.
Loading the reviews…
Took this course? Sign in to review it. Learners can write a review after they complete 25% of the lessons.