Time series analyses in practice: from data preparation to forecasting
.svg)
.svg)
.svg)
Machine Learning Engineer
1. fundamentals and explorative analysis
- Why forecasting on time series?
- Univariate and multivariate time series
- Time series components (trend, seasonality, cycles, noise)
- Visualization and descriptive statistics
2. data preprocessing and feature engineering
- Dealing with missing values and outliers
- Exogenous variables: Calendar and event characteristics
- Lag and rolling features for forecast models
3. baseline and traditional models
- Naive, seasonal and moving baselines
- Classic methods (ARIMA, ETS, Prophet)
- Probabilistic forecasting and confidence intervals
4. forecasting with deep learning methods
- Regression and ensemble models (Random Forest, XGBoost)
- Deep learning for sequences (RNN, LSTM, TCN)
- Use current frameworks: SKForecast, GluonTS, Chronos
5. evaluation and interpretability
- Use evaluation metrics (MAPE, RMSE, MASE) correctly
- Visualization of forecasts and residuals
- Explainable AI for time series models
6. deployment and monitoring
- Model export, rolling retrain, drift detection
- Best practices for the maintenance of forecast pipelines
7th Practice Challenge
- Group task: End-to-end forecast on a real database
Holistic forecasting workflow: Get to know all the steps in the data-based forecasting workflow - from the raw data set to the reliable forecast including uncertainty estimation.
Master model selection with confidence: You will learn when statistics, machine learning and deep learning deliver the greatest added value.
Future-proof architectures: Understand deployment, retrain and monitoring options to keep models stable in production.
Better business decisions: Translate forecast metrics into concrete KPIs and risk considerations.
Expandable skillset: Framework comparisons and best practices give you a roadmap for your own projects.
Support for learning transfer: cloud lab, source code, Jupyter notebooks and deploy blueprint ensure transfer to your everyday work.
This training training is conducted in a group of a maximum of 12 participants using the Zoom video conferencing software.
You work in a cloud-based lab environment provided by the trainers . All you need is a web browser and an internet connection, no software needs to be installed.
Interactive Jupyter notebooks serve as learning material and working environment. You get access to source code, documentation, references and links.
The course is held in German, the course materials are mostly in English due to the focus on programming.
There is room for your questions - individual support from the trainers is guaranteed.
You can access further materials in your personal learning environment.
Data analysts, specialists specialists, data scientists, ML engineers and business users who work with time series data and design forecast pipelines. Basic Python skills and basic knowledge of statistics are required.
.svg)
