DSC-2026-10 | Supervised Machine Learning using Python
Why is the topic important?
Machine learning is increasingly used to analyze research data, make predictions, and support decisions. Supervised learning is one of its central paradigms and is used whenever a model learns from data for which the outcome is already known. It can, for example, be applied to predict numerical measurements, classify observations, estimate the probability of an event, or identify patterns associated with a particular outcome.
Applications include understanding which factors help explain environmental variables, identifying biological or medical samples, estimating health or social outcomes, detecting known faults in technical systems, and forecasting economic indicators. Successfully applying machine learning requires understanding how models learn from data, how their performance should be evaluated, and whether their predictions are reliable and meaningful, in addition to technical programming skills such as Python.
This workshop combines the theoretical foundations of supervised learning with practical implementation in Python. Through interactive lectures and hands-on exercises, participants will learn how classical algorithms such as linear, logistic, and flexible regression models, K-Nearest Neighbours, and ensemble methods, including Random Forest, work, when to use them, and how to evaluate their performance on real-world datasets. The concepts covered also provide the foundation and baseline models needed for working critically with advanced approaches such as neural networks.
Workshop Goal
The course deals with the following topics:
- Understand the fundamental concepts and workflow of supervised learning
- Distinguish between regression and classification problems
- Understand and implement and evaluate linear, logistic, and flexible regression models for predictive modeling and binary classification tasks
- Understand and implement K-Nearest Neighbours and ensemble methods, including Random Forest
- Compare the strengths and limitations of different supervised learning algorithms and evaluate models using appropriate performance metrics
- Gain practical experience training, evaluating, and comparing models in Python
The workshop begins with an introduction to supervised learning concepts before progressing to classical machine learning algorithms for regression and classification. Participants will apply the methods to real-world datasets and learn how to select and critically evaluate suitable models for different research questions.
Workshop Content
DAY 1 Foundations of Supervised Learning:
a) Part 1: Introduction to Supervised Learning
• What is machine learning?
• Supervised vs. unsupervised learning
• Training, validation, and test sets
• Regression vs. classification problems
• Overview of the Machine Learning workflow
• Practical: Setting up the Python environment and exploring datasets
b) Part 2: Linear Regression and Linear Classification
• Intuition behind linear regression
• Cost function and least squares
• Model training and prediction
• Model evaluation metrics for regression (MAE, MSE, RMSE, R²)
• Limitations of linear relationships and introduction to Generalized Additive Models (GAMs)
• From regression to classification, linear classification concepts, and linear decision boundaries
• Practical: Implementing linear, and flexible regression and linear classification models in Python
DAY 2: Classification Algorithms and Ensemble Learning:
c) Part 3: Logistic Regression
• Limitations of linear classification
• Introduction to logistic regression
• The sigmoid function
• Probability estimation and decision thresholds
• Evaluating classification models (accuracy, precision, recall, F1-score)
• Practical: Binary classification with logistic regression in Python
d) Part 4: K-Nearest Neighbors and Ensemble Learning
• Instance-based learning
• Introduction to K-Nearest Neighbors (KNN)
• Advantages and limitations of KNN
• Introduction to ensemble learning
• Voting, Bagging, and Boosting Concepts
• Decision trees and Random Forests
• Practices of model comparison
• Practical: Comparing multiple supervised learning models in Python
Target Audience & Prior Knowledge
This workshop is intended for researchers from all disciplines who wish to build a solid foundation in supervised machine learning. Basic Python programming skills are required but can be acquired through introductory self-study materials, such as the Data Science Center’s GitHub repository “Python Introduction for Researchers”. Familiarity with basic mathematical and statistical concepts, such as equations, means, variance, and correlations, is also expected. No prior experience with machine learning models is required.
Technical Requirements
- Own laptop and connection to the Wifi (e.g. via eduroam).
- Please make sure you have access to the Jupyter4NFDI.
Maryam holds a PhD in Statistics and has extensive experience in interpretable machine learning and data science. Her research focuses on developing transparent and reliable machine learning models for real-world applications. As a data scientist at the DSC, she draws on a strong background in statistics, predictive modeling, and machine learning, as well as on extensive experience in teaching both the theoretical foundations and practical implementation of modern AI methods. Her workshops emphasize intuitive understanding, hands-on programming, and the application of machine learning techniques to real datasets.
Annika holds a master’s degree in Environmental Sciences and recently completed her PhD in global groundwater research. Her profile combines Earth system science with extensive applied programming experience in Python and R, geospatial analysis, statistics, and machine learning. In her research, she combined and analyzed large environmental datasets and applied different machine learning methods. As a data scientist at the DSC, she supports researchers with data management and analysis workflows and develops research-oriented training in data science, particularly for environmental and marine research.