Mastering Modern Time Series Forecasting: A Comprehensive Guide to Statistical, Machine Learning, and Deep Learning Models in Python

The most comprehensive guide to time series forecasting ever published — 782 pages covering every major statistical, machine learning, and deep learning model in Python.From classical ARIMA and exponential smoothing to N-BEATS, PatchTST, and foundation models, this book gives you the theory, the code, and the practical wisdom to forecast any time series.What makes this book different:Forecastability-first approach. Before building a single model, Chapter 2 teaches you to measure whether your series can be forecast — using 17 metrics no other book covers. This chapter alone is the most comprehensive treatment of forecastability in any commercial forecasting book.Every major model family, in depth. ARIMA and its variants. All 30 ETS configurations. Gradient-boosted trees (LightGBM, XGBoost, CatBoost). Deep learning architectures: LSTMs, DeepAR, N-BEATS, N-HiTS, TSMixer, TiDE, and TimeMixer. Transformer models: PatchTST, TimeXer, Crossformer, and TFT. Foundation models: Chronos, TimesFM, Moirai, and TimeGPT.Production-grade feature engineering. A 100-page chapter on feature engineering covers lag features, rolling statistics, spectral analysis, entropy measures, embedding methods, and chaos-theoretic features — with code for every technique.Rigorous performance evaluation. Two dedicated chapters on forecast metrics and state-of-the-art evaluation methodology, including proper cross-validation, statistical testing, and calibration.Implementations in Python, R, Julia, and Rust. While Python is the primary language, key models include implementations in R (forecast, fable), Julia (StateSpaceModels), and Rust (augurs, OxiDiviner) for production deployment.Who this book is for:Data scientists, ML engineers, quantitative analysts, researchers, and anyone who needs to forecast time series professionally. Assumes familiarity with Python and basic statistics.About the author:Valery Manokhin, PhD, is the creator of the most popular course on conformal prediction (5,000+ students) and author of multiple bestselling books on forecasting and machine learning. His work has been cited in academic research worldwide. Read more