Forecasting for
Data Scientists

1.01 Who am I, what is this course about?

1.02 What do we forecast? What can and can't we forecast?

1.03 Terminology and simple techniques: mean forecast, naive, random walk, SES

1.04 Forecasts are always probabilistic!, loss functions, definition of stationarity

1.05 Trend and seasonality decomposition, autocorrelation, ETS

1.06 Linear models, ARIMA, differencing

1.07 Transformations: logarithm, Box-Cox

1.08 Model selection (AIC, BIC), AutoETS, AutoARIMA

1.09 TBATS, Prophet

1.10 A brief history of forecasting competitions (M, M3, M4, CIF2016)

2.01 Global models

2.02 Machine learning methods: How to deal with non-stationarity?

2.03 Trend modelling, window-wise normalisation

2.04 How to model seasonality? Fourier terms, deseasonalisation

2.05 Normalisation

2.06 Direct vs iterative modelling, GBTs, feature engineering, ensembling, software

2.07 Deep learning for forecasting: RNNs, CNNs

2.08 Transformers for forecasting, some problems, flawed evaluations

2.09 Multivariate forecasting with deep-learning, software, data repositories, benchmarks

2.11 Newest developments: PFNs, agentic forecasting, multimodal models

3.01 Data partitioning, fixed- vs rolling origin, cross-validation, data leakage

3.02 Bias, variance, RMSE, MAE, general considerations for scale-free measures

3.03 Percentage errors, point forecasts as summary stats, MAPE, sMAPE, MAAPE, RelMAE

3.04 Calculating error measures per series and across series, WAPE/WRMSPE

3.05 MASE/RMSSE, Summary and recommendations for error measures

3.06 Probabilistic forecasting: quantile regression, simulation and bootstrapping

3.07 Evaluation of probabilistic forecasts

4.01 External regressors

4.02 Intermittent series