Thematic Course:
Machine Learning for Decision Support Systems in Oncology
Date
25-26-29 May & 3-5 June 2026
Organizer
Francesco Trovò (POLIMI)
Location
IEO, Cascina Brandezzata - Milan
About the Course
The course will cover classical Machine Learning techniques and present examples of their application in oncology. At first, it lays the foundations of ML and a broad description of the basic techniques in supervised, unsupervised, and reinforcement learning. Once the techniques have been presented, their application to some case studies is analysed, and a final hands-on experience will give the student the opportunity to apply ML techniques in real-world settings. The goals of the course are:
– Provide the students with the ability to analyze, design, and evaluate machine learning systems for decision support.
– Enable understanding of theoretical foundations and practical implementation of ML methods.
– Promote critical thinking about model performance, interpretability, and real-world impact, especially in oncology.
– Enable students to translate research or application ideas into structured ML projects.
Total number of credits: 4 credits
9:00-13:00
Lessons location: TBD
– Introduction to Machine Learning: definition, topology, and real-world examples
– One-shot support: Supervised Learning Methods
– Regression: linear regression, lasso and ridge regression, KNN, regression trees
– Classification: linear classifiers, perceptron, logistic regression, SVM, KNN, naïve Bayes
– Evaluation of supervised methods: MSE, R-square, Accuracy, precision, recall, F1-score, ROC curve
9:00-13:00
Lessons location: TBD
– Bias/Variance Tradeoff: bias-variance decomposition, feature selection, shrinkage, dimensionality reduction, bagging, boosting
– Model evaluation: validation, cross-validation, leave-one-out, adjustment techniques
– Insight retrieval: Unsupervised Learning Methods, Clustering, K-means, DBscan, PCA, ICA, visualization techniques
– Evaluation of unsupervised methods: internal and external indexes
– The NN approach: basic definition, the FFNN, other network architectures
9:00-13:00
Lessons location: TBD
– Long-Term planning: MDP: definition, examples, dynamic programming, from dynamic programming to reinforcement learning
– Reinforcement Learning: basic prediction and control methods, other control methods, challenges in applying RL to oncology
– Hands-on session: Modeling decision problems with ML, supervised, unsupervised and RL examples
9:00-13:00
Lessons location: TBD
– Explainable ML: intrinsically explainable models, coefficients understanding, Agnostic XAI methods, partial dependency plot, surrogate models
– Examples of adversarial attacks: attacks on NNs
– Oncology use case of support to decision systems using ML: immunotherapy efficacy prediction in lung cancer, behavioural recommendation with RL
9:00-13:00
Lessons location: TBD
– Examples of ML canvas: motivation and basic elements, value proposition, data, models, offline evaluation, and live monitoring
– Hands-on session: design of the ML canvas for your research, final presentation, and discussion