- Learn
- Machine Learning
- Time Series and Forecasting
ML-M11
Time Series and Forecasting
Data that arrives in time order: trend, seasonality and autocorrelation, classical models such as ARIMA and Holt-Winters, Prophet, ML and deep forecasters, and honest backtesting.
Sales per day, server load per minute, sensor readings per second: much of the data a business runs on is a time series, and its order matters. Shuffling it, as you would a table of customers, destroys exactly the patterns you need.
You will start by taking a series apart into trend, seasonality and noise and reading its autocorrelation. Then come the classical statistical models, which remain strong baselines, and Prophet for quick additive forecasts. Next you will frame forecasting as supervised learning with gradient boosting and deep models across many series at once. The module ends with backtesting: how to evaluate a forecast the way it will actually be used, without letting the future leak into training.
Lessons
- ML-11.01 Time series fundamentals A time series is a sequence of values over time, typically made of a trend, seasonal patterns, such as daily or weekly cycles, and noise. Beginner 12 min
- ML-11.02 Stationarity and classical models Many classical models assume stationarity, meaning stable statistical properties over time; differencing removes trends. Intermediate 15 min
- ML-11.03 Prophet and modern forecasting libraries Prophet models a series as trend plus seasonality plus holiday effects, fitted automatically with sensible defaults and easy-to-read components. Intermediate 10 min
- ML-11.04 ML and deep learning for forecasting Forecasting can be framed as supervised learning by using lagged values and calendar features as inputs to gradient boosting, often training one global model across thousands of related series. Advanced 14 min
- ML-11.05 Backtesting and forecast evaluation Forecasts must be evaluated the way they will be used: train on the past, predict the next period, roll forward and repeat, never letting future data leak into training. Intermediate 10 min