Time Series Forecasting
ARIMA, Prophet and gradient boosting for demand and sales forecasts.
Taught by Bibek Sharma · Data Scientist
What you'll learn
- Decompose a time series into trend, seasonal and residual components and read what each means
- Test for stationarity and apply differencing and transforms to prepare a series for modelling
- Fit, diagnose and forecast with ARIMA and SARIMA models in statsmodels
- Build a Prophet forecast with holidays, changepoints and uncertainty intervals
- Engineer lag and calendar features to forecast with LightGBM
- Backtest with rolling origins and report MAE, MAPE and MASE against a naive baseline
About this course
This time series forecasting course is for analysts, data scientists and planners who need reliable demand, sales or traffic forecasts rather than a textbook tour of theory. You will learn to decompose a series into trend, seasonality and noise, test for stationarity, and fit ARIMA and SARIMA models with statsmodels. You will then build business forecasts with Prophet, turn forecasting into a supervised learning problem with lag features and gradient boosting, and evaluate everything with rolling-origin backtesting and forecasting KPIs such as MAPE and MASE. The running example is monthly and daily retail demand, and every method ends with a forecast you can hand to a stakeholder. You should know Python and pandas and have basic statistics. By the end you will be able to choose the right forecasting method for a series, validate it honestly and explain its uncertainty.
Course content
7 lessons- 1Thinking in Time: What Makes Series DifferentWhy time-ordered data breaks ordinary ML assumptions, the components of a series, and naive baselines.Preview
- 2Decomposition: Trend, Seasonality and ResidualsSeparate a series into components with classical and STL decomposition and use them to understand the data.
- 3Stationarity, Differencing and AutocorrelationWhy ARIMA needs a stable series, how to test for it, and how ACF and PACF plots guide model order.
- 4ARIMA and SARIMA ModelsFit, diagnose and forecast with ARIMA and seasonal ARIMA in statsmodels, with automatic order selection.
- 5Prophet for Business ForecastsFit Prophet with holidays, changepoints and regressors, and produce forecasts stakeholders can read.
- 6Forecasting With Lag Features and Gradient BoostingReframe forecasting as supervised learning with lags, rolling stats and calendar features, then train LightGBM.
- 7Backtesting and Forecasting KPIsRolling-origin evaluation, the metrics that matter for planning, and how to report forecast quality.