Examples and Tutorials¶
Practical examples and tutorials to help you understand and apply skforecast. You can also find them in the free track Forecasting with Python.
Skforecast ecosystem¶
-
Agentic forecasting with skforecast-AI
An AI assistant that profiles your series, picks and validates a model, and returns the runnable skforecast script behind the forecast.
-
skforecast-AI MCP server for coding agents
Let Claude Code, Cursor, VS Code or Codex call skforecast-AI as forecasting tools through the Model Context Protocol.
-
Skforecast Studio: time series forecasting made simple
Build forecasting models visually, no code required, and export production-ready skforecast Python code.
Machine learning forecasting¶
-
Skforecast: time series forecasting with machine learning
Build your first machine learning forecaster end to end, from lags to backtesting.
-
Forecasting with gradient boosting: XGBoost, LightGBM and CatBoost
Compare the three main gradient boosting libraries on the same forecasting problem.
-
A focused guide to using XGBoost as the estimator inside a forecaster.
-
A focused guide to using LightGBM as the estimator inside a forecaster.
-
Modelling time series trend with tree based models
Why trees cannot extrapolate a trend, and the differentiation and detrending that fix it.
-
Forecasting of categorical time series
Predict the next category instead of the next value, using a classifier as the estimator.
-
Interpretable forecasting models
Explain what drives a forecast using feature importance and SHAP values.
-
Stacking ensemble of machine learning models to improve forecasting
Combine several forecasters into a meta-model that beats each one alone.
Global models: multi-series forecasting¶
-
Global Forecasting Models I: Multi-series forecasting
Train a single model on many series at once and predict them all.
-
Global Forecasting Models II: Comparative analysis of single and multi-series forecasting
When does one global model beat one model per series? A side-by-side comparison.
-
Global Forecasting Models III: Modeling thousand time series with a single global model
Scale to thousands of series while keeping training time and memory under control.
-
Global Forecasting Models IV: A step by step guide using Kaggle sticker sales data
A complete walkthrough of a Kaggle forecasting competition with a global model.
-
Global Forecasting Models V: The M5 accuracy competition
What the winning M5 solutions teach us about global forecasting models.
-
Clustering time series to improve forecasting
Group similar series first, then fit one global model per cluster.
Foundation models and deep learning¶
-
Forecasting with foundation models
Zero-shot forecasting with pre-trained models such as Chronos, TimesFM and Moirai.
-
Data leakage in pre-trained forecasting models
How pre-training data can contaminate a zero-shot benchmark, and how to check for it.
-
Forecasting with deep learning
RNN and LSTM forecasters with ForecasterRnn, from architecture to predictions.
-
Forecasting at scale with deep learning
Train a single neural network over many series without losing per-series accuracy.
Statistical models¶
-
Forecasting with statistical models
Overview of the classical statistical models available through ForecasterStats.
-
Fit, tune and validate ARIMA and SARIMAX models, including automatic order selection.
-
ETS and Holt-Winters models for series with trend and seasonality.
-
ARAR forecasting models in Python
An automated alternative to ARIMA that shortens long memory before modelling the remaining signal.
Probabilistic forecasting¶
-
Probabilistic forecasting with machine learning
Prediction intervals and quantiles via bootstrapping, conformal methods and quantile regression.
-
Probabilistic forecasting: prediction intervals for multi-step forecasting
Prediction intervals for multi-step horizons, balancing coverage against interval width.
-
Continuous Ranked Probability Score (CRPS) in probabilistic forecasting
Evaluate probabilistic forecasts with CRPS, the metric that scores the whole predicted distribution.
Data exploration, quality and monitoring¶
-
Plots that reveal trend, seasonality and autocorrelation before you model anything.
-
Forecasting time series with missing values
Strategies for gaps and NaNs, from imputation to NaN-tolerant estimators.
-
Intermittent demand forecasting
Forecasting sparse series that are zero most of the time.
-
Mitigating the impact of covid on forecasting models
Down-weight or exclude anomalous periods with weighted time series forecasting.
-
Use forecast errors and prediction intervals to flag anomalous observations.
-
Data drift detection in time series forecasting
Spot when production data has moved away from what the model was trained on.
Real-world case studies¶
-
Forecasting energy demand with machine learning
Hourly electricity demand with calendar and weather exogenous variables.
-
Forecasting web traffic with machine learning and Python
Daily visits to a website, including weekly seasonality and special events.
-
Bitcoin price prediction with Python
A cryptocurrency case study, and an honest look at how well it can work.
Performance¶
-
Accelerate forecasting models with GPU
Move gradient boosting training onto a GPU and measure what you actually gain.
-
Accelerate forecasting models with Intel scikit-learn-intelex
Speed up scikit-learn estimators on Intel CPUs without changing your code.