Time Series Hub – Open Datasets & Models for Forecasting and Analysis

Time Series Hub is a community-driven initiative dedicated to advancing time series analysis through open datasets, state‑of‑the‑art models, and educational resources. We bring together researchers, practitioners, and enthusiasts working on forecasting, anomaly detection, classification, and explainability for temporal data across domains like finance, weather, energy, healthcare, and IoT.

---

🎯 Our Mission

• Curate and share high‑quality open datasets for time series tasks (univariate, multivariate, irregular, long‑range).
• Develop and benchmark cutting‑edge models – from classical statistical methods (ARIMA, ETS) to deep learning (LSTM, Transformers, TCNs) and foundation models (TimesFM, Chronos, Moirai).
• Provide reproducible pipelines for data preprocessing, model training, and evaluation.
• Foster a collaborative community around time series machine learning through tutorials, competitions, and open‑source contributions.

---

🚀 Interactive Demos

Explore our live Hugging Face Spaces and experiment with time series models directly in your browser:

🔮 Forecasting

• Univariate Forecasting Playground – Try ARIMA, Prophet, and DeepAR on your own data.
• Multivariate Forecasting – Predict multiple related time series (e.g., weather variables) with Transformers.
• Probabilistic Forecasting – Generate prediction intervals and quantile forecasts.

⚠️ Anomaly Detection

• Anomaly Detection Demo – Detect outliers in sensor data using LSTM autoencoders and Isolation Forest.
• Real‑time Anomaly Detection – Simulate streaming data and spot anomalies on the fly.

📊 Visualization & Analysis

• Interactive Time Series Explorer – Upload CSV and visualize trends, seasonality, and residuals.
• Seasonality Decomposition – Decompose time series into trend, seasonal, and residual components.
• Cross‑Correlation Analysis – Explore lagged relationships between multiple series.

🧠 Model Interpretability

• SHAP for Time Series – Explain predictions of black‑box forecasting models.
• Attention Visualization – See which time steps a Transformer model focuses on.

All demos support CSV upload – no installation required!

---

🧠 Research Focus Areas

📈 Forecasting

• Long‑range forecasting with Transformers (Informer, Autoformer, PatchTST).
• Probabilistic forecasting (DeepAR, GP, Normalizing Flows).
• Hierarchical and grouped time series forecasting.

🛡️ Anomaly Detection

• Unsupervised and semi‑supervised methods for detecting anomalies in time series.
• Real‑time anomaly detection in streaming data.
• Explainable anomaly detection for critical applications (finance, healthcare).

🌦️ Domain‑Specific Applications

• Weather Forecasting – Medium‑range and subseasonal forecasting using physical + ML models.
• Energy & Load Forecasting – Smart grid, renewable energy prediction.
• Financial Time Series – Volatility forecasting, algorithmic trading signals.
• Healthcare – Patient vital signs, epidemic spread modeling.

🤖 Foundation Models for Time Series

• Leveraging large‑scale pretrained models (TimesFM, Chronos, Lag‑Llama) for zero‑shot forecasting.
• Fine‑tuning foundation models on domain‑specific data.
• Evaluating generalization across datasets and domains.

📊 Benchmarks & Evaluation

• Standardized benchmarks (Monash Time Series Forecasting Repository, ETT, M4, M5).
• Metrics for point forecasts, probabilistic forecasts, and anomaly detection (MASE, CRPS, F1, etc.).
• Reproducibility frameworks and leaderboards.

---

📦 Models & Datasets

We host and link to open time series resources on Hugging Face Hub.

We are continuously adding new models and datasets. Follow our organization page for updates.

---

📚 Educational Resources

Learn time series analysis and forecasting with our free tutorials and materials.

• Interactive Notebooks – Step‑by‑step guides: from ARIMA to Transformers in Python.
• Video Lectures – Recorded talks on time series fundamentals, deep learning, and applications.
• Course Materials – Slides, exercises, and projects from university‑level courses.
• Blog Posts – Deep dives into new models, datasets, and techniques.

---

📝 Selected Publications

1. "PatchTST: A Time Series is Worth 64 Words" – ICLR 2023
2. "Autoformer: Decomposition Transformers with Auto‑Correlation for Long‑Term Series Forecasting" – NeurIPS 2021
3. "Informer: Beyond Efficient Transformer for Long Sequence Time‑Series Forecasting" – AAAI 2021
4. "TimesFM: A Foundation Model for Time Series Forecasting" – Google Research 2024
5. "Chronos: Learning the Language of Time Series" – Amazon Science 2024

Full list with links to PDFs available on our Publications Page.

---

🤝 Get Involved

We welcome contributions from the community – whether you are a researcher, engineer, student, or domain expert.

For Researchers

• Share your datasets and models on our Hub.
• Collaborate on benchmarks and comparative studies.
• Propose new challenges and tasks.

For Developers

• Integrate our models into your applications.
• Report bugs or suggest improvements via GitHub.
• Contribute to our open‑source codebase.

For Practitioners

• Use our demos and models for real‑world problems.
• Provide feedback on model performance and usability.
• Share use cases and success stories.

For Students

• Learn with our tutorials and notebooks.
• Participate in forecasting competitions.
• Start your own research project with our support.

---

🌐 Connect With Us

• 🤗 Hugging Face: TimeSeriesHub – Models, datasets, and spaces.
• 💻 GitHub: TimeSeriesHub – Source code, development, and issue tracking.
• 📧 Email: contact@timeserieshub.org – General inquiries and collaboration.
• 📝 Blog: Medium/TimeSeriesHub – In‑depth articles.

---

🔄 Ecosystem Integration

Our work integrates with the broader Hugging Face and scientific Python ecosystems:

• Models on the Hub with easy‑to‑use transformers‑style APIs.
• Datasets with streaming and preprocessing pipelines.
• Spaces for interactive demos built with Gradio and Plotly.
• Gradio apps for user‑friendly interfaces.
• Interoperability with libraries like sktime, darts, statsmodels, and PyTorch Forecasting.

---

Empowering time series analysis through open collaboration and state‑of‑the‑art tools.