Advanced machine learning time series....
summary
Survey of modern time-series methods with concrete claims: Temporal Fusion Transformer and hybrid architectures (ES-Hybrid, N-BEATS, Tsetlin) repeatedly beat classical stats baselines and win benchmarks, sometimes improving scores by ~50% versus single models. Open-source frameworks and ready-to-run libraries are shown, experiments include hyperparameter sweeps, feature selection that flags the 168-hour–7-day lag as most informative, and end-to-end pipelines that train in cloud and yield reproducible probabilistic forecasts.
AI-written summary · about
chapters (10)
- 0:00Introduction to time series forecasting
- 3:19Breakthrough in machine learning models
- 6:59Hybrid models and frameworks
- 10:35Business applications and accuracy
- 13:27Data preprocessing and transformations
- 17:53Statistical vs neural network models
- 20:41Transformer models and advantages
- 24:13Hyperparameter optimization
- 26:51Performance improvements and experiments
- 30:58Summary and future directions
related talks
39mForecasting With Classical and Machine Learning Methods
43mFeature Engineering for Time Series Forecasting
31mFoundational Time Series Models in Practice: The Future of Forecasting, or Just Hype?
54mTransformes for Time Series: Is the New State of the Art (SOA) Approaching?
31mML time series forecasting methods
19mForecasting for Everyone - Foundation Models and the Future of Time Series Prediction
31mFoundational Models for Time Series Forecasting: are we there yet?
01h 27mIntroduction to Machine Learning with Time Series