Forecasting and time series

About this course

This course covers time series forecasting: how to frame a forecasting task, honestly validate a model against historical data, and avoid fooling yourself with a rolling backtest. It solves the gap between "the model produced a nice-looking chart" and a forecast you can actually rely on at work.

The course moves from naive baselines and time-based splits through seasonality, trend, and outliers to statistical models, feature-based models, and forecast intervals. Projects run on your own series: build an honest validation rig with a protocol and metrics, compile a structural profile of your series, defend a forecasting pipeline with naive baselines and your chosen model.

Built for analysts and developers who have their own series — sales, support tickets, load — and need to forecast it. You'll need basic statistics and Python; prior time series experience isn't required.

What you'll learn

• Frame a forecasting task: object, granularity, horizon, decision
• Build naive baselines and explain why they're hard to beat
• Run a rolling backtest and avoid leakage from the future
• Measure forecast error in decision units, not percentages
• Diagnose trend, seasonality, and outliers in your own series
• Build forecast intervals instead of relying on a point estimate
• Compare statistical and feature-based models on the same window
• Account for forecast bias and human adjustments in reporting

What you'll end up with

• A validation rig with a protocol and metrics built on your own data
• A structural profile of your series, covering seasonality and outliers
• A defended forecasting pipeline with baselines and your chosen model
• A certificate on paid plans

Not for you if you want one universal forecasting algorithm without digging into your own data and validating it on your own series.