Time series, Analytics and Forecasting - Course Presentation

L1

Time series - starter pack and visualization

• Introduction - What can be forecasted? • Time series visualization • Embeddings • Diagnostic plots for point and probabilistic forecasts

Time series graphics & review of Fourier decomposition

Forecasting with Fourier transform

L2

What can be forecasted

• correlation, ACF and random walks, stationarity and forecastability spectral, approximate, sample entropy. • static benchmark models (persistent, conditional mean), residual analysis, linear

Naive models coding

Benchmark models and bootstrapped scenarios

L3

Stationarity and forecasting assumptions

• STL decompositions, spectral analysis • differentiation, log, box-cox transform • removing closed-loop effects • anomaly detection in TS • data imputation for TS

STL decomposition for forecasting

Outlier detection methods: boxplots, sigma filters, conditional means, forecasting

L4

State Space models

• Exponential smoothing family • Variance computation of SS models under NID assumption • connection between SS models and stateless models

SS models and input-output maps - ES implementation

mini-challenge: single time series forecasting pipeline

L5

Hierarchical forecasting

• bottom up • optimal reconciliation • probabilistic hierarchical forecasting

Introducing the final project

flipped classroom preparation

L6

Advanced forecasting models

• AR, ARIMA, ARIMAX advanced models (random forests, gradient boosting, NNs for forecasting) • case studies: amazon and uber forecasting pipelines

ARX implementation and comparison with advanced models

flipped classroom: presentation of an advanced topic Hierarchical forecasting

L7

Probabilistic forecasting & scenario generation without NID assumption

• quantile regression, conformal predictions, bootstrap • scenario generation for single and multi-output models

Conformal prediction vs bootstrap

final project

L8

TS model selection and validation

• dealing with data scarcity • CV for time series and hyperpar tuning, examples of spillage • Diebold-Mariano and Nemenyi tests. Bonferroni correction

Purged k-fold CV

final project

L9

Decision under uncertainty

• single stage decisions • scenario optimization

Maximum force off times for a heat pump

final project

L10

Decision under uncertainty

• stochastic tree optimization

stochastic control of a battery

final project