ID:
510488
Durata (ore):
66
CFU:
9
SSD:
Statistica
Anno:
2026
Dati Generali
Periodo di attività
Primo Semestre (21/09/2026 - 18/12/2026)
Syllabus
Obiettivi Formativi
By the end of the course, students will be able to:
• understand the fundamental principles of statistical learning;
• distinguish between prediction and inference;
• understand supervised and unsupervised learning problems;
• explain the concepts of training error, test error, overfitting, bias, variance, and generalization;
• implement the main statistical learning methods in R;
• construct appropriate training, validation, and test workflows;
• apply cross-validation and other resampling procedures;
• perform regularization and model selection;
• compare alternative models using appropriate performance metrics;
• recognize and avoid common sources of data leakage;
• critically interpret regression, classification, dimension-reduction, and clustering results;
• evaluate the trade-off between predictive performance and interpretability;
• identify weaknesses in statistical learning pipelines;
• use generative AI tools responsibly for coding and statistical analysis;
• critically assess AI-generated code, model choices, and interpretations;
• communicate the results of a statistical learning analysis clearly and effectively.
• understand the fundamental principles of statistical learning;
• distinguish between prediction and inference;
• understand supervised and unsupervised learning problems;
• explain the concepts of training error, test error, overfitting, bias, variance, and generalization;
• implement the main statistical learning methods in R;
• construct appropriate training, validation, and test workflows;
• apply cross-validation and other resampling procedures;
• perform regularization and model selection;
• compare alternative models using appropriate performance metrics;
• recognize and avoid common sources of data leakage;
• critically interpret regression, classification, dimension-reduction, and clustering results;
• evaluate the trade-off between predictive performance and interpretability;
• identify weaknesses in statistical learning pipelines;
• use generative AI tools responsibly for coding and statistical analysis;
• critically assess AI-generated code, model choices, and interpretations;
• communicate the results of a statistical learning analysis clearly and effectively.
Prerequisiti
Knowledge of basic concepts of statistics: confidence intervals, test, p-values
Metodi didattici
The course consists of 33 sessions of approximately 90 minutes each.
The course includes:
• 18 theory lectures
• 15 coding and applied laboratory sessions
Coding sessions are conducted in RStudio (download R first, at https://cran.mirror.garr.it/CRAN/ and then RStudio at https://posit.co/downloads).
The course combines:
• theoretical lectures;
• guided coding sessions;
• live demonstrations;
• statistical problem solving;
• model comparison;
• debugging exercises;
• discussion of common analytical mistakes;
• AI-assisted analysis and critical evaluation.
The emphasis is not only on learning how to fit statistical learning models, but on understanding when a model is appropriate, how it should be validated and whether its results can be trusted.
The course includes:
• 18 theory lectures
• 15 coding and applied laboratory sessions
Coding sessions are conducted in RStudio (download R first, at https://cran.mirror.garr.it/CRAN/ and then RStudio at https://posit.co/downloads).
The course combines:
• theoretical lectures;
• guided coding sessions;
• live demonstrations;
• statistical problem solving;
• model comparison;
• debugging exercises;
• discussion of common analytical mistakes;
• AI-assisted analysis and critical evaluation.
The emphasis is not only on learning how to fit statistical learning models, but on understanding when a model is appropriate, how it should be validated and whether its results can be trusted.
Verifica Apprendimento
The final grade is based entirely on the final examination:
Data analysis project (no live presentation is requested): 30%
Statistical and methodological defence: 35%
Live R modification / debugging: 35%
Students must demonstrate individual understanding of all methodological choices, code, results, and interpretations included in the submitted project.
Data analysis project (no live presentation is requested): 30%
Statistical and methodological defence: 35%
Live R modification / debugging: 35%
Students must demonstrate individual understanding of all methodological choices, code, results, and interpretations included in the submitted project.
Testi
James, G., Witten, D., Hastie, T., Tibshirani, R. and Taylor, J.
An Introduction to Statistical Learning: with Applications in R.
Springer.
Hastie, T., Tibshirani, R. and Friedman, J.
The Elements of Statistical Learning.
Springer.
Additional readings, R materials, datasets, and coding exercises will be provided during the course.
An Introduction to Statistical Learning: with Applications in R.
Springer.
Hastie, T., Tibshirani, R. and Friedman, J.
The Elements of Statistical Learning.
Springer.
Additional readings, R materials, datasets, and coding exercises will be provided during the course.
Contenuti
Part I — Foundations of Statistical Learning
Lecture 1 — Introduction to Statistical Learning
Statistical learning; prediction versus inference; supervised/unsupervised and parametric/nonparametric methods; training/test error; generative AI.
Lab 1 — R and RStudio Fundamentals
R/RStudio; objects and assignment; vectors, matrices, lists, data frames, factors; missing values; functions, control structures and data generation.
Lecture 2 — The Statistical Learning Framework
Generalization; reducible/irreducible error; complexity; overfitting/underfitting; bias–variance trade-off; loss functions.
Lab 2 — Data Exploration and Visualization
Data import, inspection and dplyr manipulation; descriptive statistics; missing values; exploration; ggplot2; anomalies.
Part II — Regression and Model Validation
Lecture 3 — Linear Regression I
Simple and multiple linear regression; least-squares estimation; interpretation of regression coefficients; prediction.
Lecture 4 — Linear Regression II
Categorical predictors; interactions; nonlinear transformations; regression diagnostics and residual analysis; model comparison.
Lab 3 — Linear Regression in R
lm(); coefficients; confidence/prediction intervals; multiple regression, interactions, polynomial terms; residual diagnostics; prediction on new data.
Lecture 5 — Model Assessment and Validation
Training, validation and test sets; validation-set approach; leave-one-out, k-fold and repeated cross-validation; bootstrap; model selection and generalization error.
Lab 4 — Cross-Validation and Reproducible Workflows
Data splitting; reproducible randomization; tidymodels, recipes and workflows; resampling; cross-validation; preprocessing and data leakage.
Lecture 6 — Regularization and High-Dimensional Regression
Subset selection; shrinkage; Ridge regression and Lasso; tuning parameter selection; variable selection.
Lab 5 — Ridge Regression and Lasso
glmnet; standardization; cross-validation for penalty selection; Ridge versus Lasso; coefficient paths; variable selection; Elastic Net.
Part III — Classification
Lecture 7 — Logistic Regression
Binary classification; probabilities, odds and log-odds; logistic function; maximum likelihood estimation; classification thresholds.
Lecture 8 — Classification Performance
Confusion matrix; accuracy; sensitivity and specificity; precision and recall; F1 score; ROC curves and AUC; classification thresholds; class imbalance.
Lab 6 — Logistic Regression and Classification
Logistic regression in R; probability/class prediction; confusion matrix; sensitivity, specificity, precision, F1; ROC/AUC; thresholds; class imbalance.
Lecture 9 — Model Selection and Resampling
Selection versus assessment; training/validation/test sets; model comparison through common and repeated resampling; performance variability; baselines; selection bias; test-set contamination.
Lecture 10 — Hyperparameter Tuning and Model Comparison
Hyperparameters versus estimated parameters; complexity; grid search; cross-validation for tuning; nested preprocessing/tuning/validation logic; selection bias; alternative specifications; final selection and untouched test-set evaluation.
Lab 7 — Model Selection and Tuning in R
Train/validation/test logic; common k-fold and repeated cross-validation; model specifications; parameters/hyperparameters; tidymodels grid search; results visualization; workflow finalization; training-set fitting; untouched test evaluation; variability; preprocessing within folds; leakage prevention.
Part IV — Tree-Based and Nonlinear Methods
Lecture 11 — Decision Trees
Regression and classification trees; recursive binary splitting; impurity measures; tree depth; pruning; interpretability.
Lab 8 — Decision Trees
Tree fitting, visualization and prediction; cost-complexity pruning; hyperparameter tuning; overfitting in trees.
Lecture 12 — Bagging and Random Forests
Bootstrap aggregation and bagging; random feature selection; random forests; out-of-bag error; variable importance.
Lecture 13 — Boosting
Sequential learning and weak learners; gradient boosting; number of trees, tree depth and learning rate; overfitting and regularization.
Lab 9 — Random Forests and Boosting
Random forest implementation; hyperparameter tuning; variable importance; gradient boosting; model comparison; predictive accuracy versus interpretability.
Lecture 14 — Predictive Modelling and Critical Evaluation
Linear/nonlinear and simple/flexible models; performance versus interpretability; overfitting; tuning sensitivity; stability; importance limitations; robustness; final selection; limits of complexity; critical interpretation.
Lab 10 — Model Comparison and Robustness in R
Common workflows, baselines, splits and folds; logistic regression, trees, forests and boosting; training-only tuning; cross-validation and test evaluation; variability; accuracy/complexity; importance; sensitivity to splits/seeds; overfitting, stability, robustness; final selection.
Part V — Unsupervised Learning
Lecture 15 — Principal Component Analysis
Dimension reduction; variance decomposition; principal components; loadings and scores; explained variance; interpretation; standardization.
Lab 11 — PCA and Clustering, Part I
PCA in R; scree plots; component scores and loadings; biplots; interpretation of principal components.
Lecture 16 — Clustering
Unsupervised learning; similarity and distance; K-means and hierarchical clustering; linkage criteria; dendrograms; choice of the number of clusters; cluster validation.
Lab 11 — PCA and Clustering, Part II
K-means; elbow method; silhouette analysis; hierarchical clustering; complete, single, average and Ward linkage; comparison of clustering solutions.
Part VI — Interpretation, Robustness and AI-Assisted Learning
Lecture 17 — Model Interpretation and Explainability
Global and local interpretation; variable importance; partial dependence and Individual Conditional Expectation (ICE); model complexity and interpretability; prediction versus causation; limits and instability of explanations.
Lecture 18 — Leakage, Robustness, AI and Course Integration
Leakage; preprocessing/validation errors; selection bias; tuning errors; baselines; fair comparison; variability, stability, robustness; reproducibility; reporting; AI assistance, limitations, hallucinations and code validation; human responsibility; integrated workflows and final project.
Lab 12 — Debugging and AI-Assisted Learning
Programming/statistical errors; leakage; misleading metrics; test contamination; invalid comparisons; reproducibility; AI-generated R code audits; workflow correction; valid alternatives; AI-use documentation; critical interpretation.
Lab 13 — Model Interpretation and Explainability in R
Performance and interpretation; global/permutation importance; partial dependence/ICE; local explanations; forest predictions; correlated predictors and importance instability; predictive versus causal interpretation.
Lab 14 — Data Leakage, Robustness and Model Criticism in R
Preprocessing leakage; training/test performance; training-only validation/tuning; test contamination; resampling variability; misleading accuracy under imbalance; temporal, target and feature-selection leakage; final test evaluation.
Lab 15 — AI-Assisted Learning and Final Challenge
AI-generated R analysis audits; train/validation/test workflows; baselines; fair comparison of logistic regression, trees, forests and boosting; common folds/metrics; training-only tuning; final test evaluation; interpretation; robustness; human/AI responsibility; defending a trustworthy workflow.
Lecture 1 — Introduction to Statistical Learning
Statistical learning; prediction versus inference; supervised/unsupervised and parametric/nonparametric methods; training/test error; generative AI.
Lab 1 — R and RStudio Fundamentals
R/RStudio; objects and assignment; vectors, matrices, lists, data frames, factors; missing values; functions, control structures and data generation.
Lecture 2 — The Statistical Learning Framework
Generalization; reducible/irreducible error; complexity; overfitting/underfitting; bias–variance trade-off; loss functions.
Lab 2 — Data Exploration and Visualization
Data import, inspection and dplyr manipulation; descriptive statistics; missing values; exploration; ggplot2; anomalies.
Part II — Regression and Model Validation
Lecture 3 — Linear Regression I
Simple and multiple linear regression; least-squares estimation; interpretation of regression coefficients; prediction.
Lecture 4 — Linear Regression II
Categorical predictors; interactions; nonlinear transformations; regression diagnostics and residual analysis; model comparison.
Lab 3 — Linear Regression in R
lm(); coefficients; confidence/prediction intervals; multiple regression, interactions, polynomial terms; residual diagnostics; prediction on new data.
Lecture 5 — Model Assessment and Validation
Training, validation and test sets; validation-set approach; leave-one-out, k-fold and repeated cross-validation; bootstrap; model selection and generalization error.
Lab 4 — Cross-Validation and Reproducible Workflows
Data splitting; reproducible randomization; tidymodels, recipes and workflows; resampling; cross-validation; preprocessing and data leakage.
Lecture 6 — Regularization and High-Dimensional Regression
Subset selection; shrinkage; Ridge regression and Lasso; tuning parameter selection; variable selection.
Lab 5 — Ridge Regression and Lasso
glmnet; standardization; cross-validation for penalty selection; Ridge versus Lasso; coefficient paths; variable selection; Elastic Net.
Part III — Classification
Lecture 7 — Logistic Regression
Binary classification; probabilities, odds and log-odds; logistic function; maximum likelihood estimation; classification thresholds.
Lecture 8 — Classification Performance
Confusion matrix; accuracy; sensitivity and specificity; precision and recall; F1 score; ROC curves and AUC; classification thresholds; class imbalance.
Lab 6 — Logistic Regression and Classification
Logistic regression in R; probability/class prediction; confusion matrix; sensitivity, specificity, precision, F1; ROC/AUC; thresholds; class imbalance.
Lecture 9 — Model Selection and Resampling
Selection versus assessment; training/validation/test sets; model comparison through common and repeated resampling; performance variability; baselines; selection bias; test-set contamination.
Lecture 10 — Hyperparameter Tuning and Model Comparison
Hyperparameters versus estimated parameters; complexity; grid search; cross-validation for tuning; nested preprocessing/tuning/validation logic; selection bias; alternative specifications; final selection and untouched test-set evaluation.
Lab 7 — Model Selection and Tuning in R
Train/validation/test logic; common k-fold and repeated cross-validation; model specifications; parameters/hyperparameters; tidymodels grid search; results visualization; workflow finalization; training-set fitting; untouched test evaluation; variability; preprocessing within folds; leakage prevention.
Part IV — Tree-Based and Nonlinear Methods
Lecture 11 — Decision Trees
Regression and classification trees; recursive binary splitting; impurity measures; tree depth; pruning; interpretability.
Lab 8 — Decision Trees
Tree fitting, visualization and prediction; cost-complexity pruning; hyperparameter tuning; overfitting in trees.
Lecture 12 — Bagging and Random Forests
Bootstrap aggregation and bagging; random feature selection; random forests; out-of-bag error; variable importance.
Lecture 13 — Boosting
Sequential learning and weak learners; gradient boosting; number of trees, tree depth and learning rate; overfitting and regularization.
Lab 9 — Random Forests and Boosting
Random forest implementation; hyperparameter tuning; variable importance; gradient boosting; model comparison; predictive accuracy versus interpretability.
Lecture 14 — Predictive Modelling and Critical Evaluation
Linear/nonlinear and simple/flexible models; performance versus interpretability; overfitting; tuning sensitivity; stability; importance limitations; robustness; final selection; limits of complexity; critical interpretation.
Lab 10 — Model Comparison and Robustness in R
Common workflows, baselines, splits and folds; logistic regression, trees, forests and boosting; training-only tuning; cross-validation and test evaluation; variability; accuracy/complexity; importance; sensitivity to splits/seeds; overfitting, stability, robustness; final selection.
Part V — Unsupervised Learning
Lecture 15 — Principal Component Analysis
Dimension reduction; variance decomposition; principal components; loadings and scores; explained variance; interpretation; standardization.
Lab 11 — PCA and Clustering, Part I
PCA in R; scree plots; component scores and loadings; biplots; interpretation of principal components.
Lecture 16 — Clustering
Unsupervised learning; similarity and distance; K-means and hierarchical clustering; linkage criteria; dendrograms; choice of the number of clusters; cluster validation.
Lab 11 — PCA and Clustering, Part II
K-means; elbow method; silhouette analysis; hierarchical clustering; complete, single, average and Ward linkage; comparison of clustering solutions.
Part VI — Interpretation, Robustness and AI-Assisted Learning
Lecture 17 — Model Interpretation and Explainability
Global and local interpretation; variable importance; partial dependence and Individual Conditional Expectation (ICE); model complexity and interpretability; prediction versus causation; limits and instability of explanations.
Lecture 18 — Leakage, Robustness, AI and Course Integration
Leakage; preprocessing/validation errors; selection bias; tuning errors; baselines; fair comparison; variability, stability, robustness; reproducibility; reporting; AI assistance, limitations, hallucinations and code validation; human responsibility; integrated workflows and final project.
Lab 12 — Debugging and AI-Assisted Learning
Programming/statistical errors; leakage; misleading metrics; test contamination; invalid comparisons; reproducibility; AI-generated R code audits; workflow correction; valid alternatives; AI-use documentation; critical interpretation.
Lab 13 — Model Interpretation and Explainability in R
Performance and interpretation; global/permutation importance; partial dependence/ICE; local explanations; forest predictions; correlated predictors and importance instability; predictive versus causal interpretation.
Lab 14 — Data Leakage, Robustness and Model Criticism in R
Preprocessing leakage; training/test performance; training-only validation/tuning; test contamination; resampling variability; misleading accuracy under imbalance; temporal, target and feature-selection leakage; final test evaluation.
Lab 15 — AI-Assisted Learning and Final Challenge
AI-generated R analysis audits; train/validation/test workflows; baselines; fair comparison of logistic regression, trees, forests and boosting; common folds/metrics; training-only tuning; final test evaluation; interpretation; robustness; human/AI responsibility; defending a trustworthy workflow.
Lingua Insegnamento
INGLESE
Corsi
Corsi
INTERNATIONAL BUSINESS AND ENTREPRENEURSHIP
Laurea Magistrale
2 anni
No Results Found
Persone
Persone
No Results Found