Get better insights from time-series data and become proficient in building models with real-world data
Key Features:
- Explore time series forecasting and time series analysis in Python using ARIMA, SARIMA, GARCH, gradient boosting, and recurrent neural networks.
- Improve predictive modeling with feature engineering and forecasting machine learning techniques.
- Apply demand forecasting and financial forecasting methods through practical case studies and real-world datasets.
Book Description:
Machine Learning for Time Series, Second Edition offers a practical guide to building forecasting systems with the most important libraries such as sktime, darts, Nixtla's skforecast and mlforecast, and AutoGluon among others. Starting with the fundamentals, you'll learn how to characterize datasets, measure and validate prediction performance, engineer features, and build forecasting pipelines.
The book covers classical methods such as ARIMA and Theta, machine learning methods such as gradient boosting, and deep learning with recurrent, latest Transformer-based (like Crossformer), or convolutional (TimesNet) models. We'll apply foundation models such as TimesFM, Chronos, and OpenTSLM, or generative approaches like Chain-of-Charts, and uncertainty quantification with conformal prediction such as GARCH conformal, conformalized quantile regression, or split conformal. We'll also dive into production infrastructure like feature stores, model tracking, and monitoring and scaling strategies, as well as classification, self-supervised encoding, spatiotemporal modeling, handling irregular time series with ODEs, survival analysis, and causal inference.
Through practical examples, you'll learn how to evaluate models, detect and handle anomalies and drift, and choose approaches that match data and operational needs.
What You Will Learn:
- Apply classical time series models such as ETS, ARIMA, and Theta
- Validate with SMAPE, MASE, RMSSE and rolling window, and walk-forward
- Forecast as regression with sktime, darts, mlforecast, and skforecast
- Use modern deep learning like LSTMs, DLinear, TSMixer, and TimesNet
- Grasp foundation models such as TimesFM, OpenTSLM, or Chain-of-Charts
- Quantify uncertainty quantification with conformalized regression
- Set up feature stores, MLflow, and monitoring and scaling strategies
- Handle classification, drift, spatiotemporal and irregular time series
Who this book is for:
This book is ideal for data analysts, data scientists, and Python developers who want instantly useful and practical recipes to implement today, and a comprehensive reference book for tomorrow. Basic knowledge of the Python Programming language is a must, while familiarity with statistics will help you get the most out of this book.
Table of Contents
- Towards Modern Forecasting
- Preparing and Visualizing Time Series Data
- Classical Models and Validation
- Forecasting with Machine Learning
- Feature Engineering and Tree-Based Models
- Multivariate and Hierarchical Forecasting
- Practical Deep Learning for Time Series
- Quantifying Time Series Uncertainty with Conformal Prediction
- Foundation Models: Quantitative and Qualitative Forecasting
- Production Workflows: Deployment, Monitoring, and Scaling
- Beyond Forecasting: Specialized Applications
- Intermittent Forecasting and Survival Analysis
