On February 13^th, 2023 at 11.50 am Antonio Consolo, PHD Student in Information Technology, will give a seminar on "Sparse randomized decision trees for classification and regression" in DEIB Conference Room.

In Machine Learning and Statistics decision trees are widely used for classification and regression tasks arising in a variety of application fields. They have a flowchart-like structure where each branch node represents a ”test” on predictor variables and each leaf node represents either a discrete decision (e.g. a class label) or a continuous response. Since designing optimal decision trees is NP-hard, all classical methods are based on simple greedy approaches. Due to the remarkable progress in the computational performance of Mixed-Integer Linear Programming and nonlinear optimization solvers, decision trees have been recently revisited. The growing attention they have attracted during the past years is also due to their interpretability and also the ability to induce fairness measures. The goal of my project is to investigate a continuous optimization formulation to build sparse randomized decision trees. In this models a random decision is made at each decision node of the tree with a certain probability. The probabilistic nature of this model gives us more flexibility and additional information in terms of the posterior probability.