Hybrid seminar at GERAD and on Zoom .

We study multi-attribute decision-making and binary classification problems, develop interpretable threshold-based classification rules, and propose direct and noisy inverse optimization frameworks for learning these rules from observed classifications. We establish the theoretical foundations of the proposed approaches and benchmark their performance against several machine learning classifiers, including random forests, support vector machines, logistic regression, ridge classifiers, decision trees, and neural networks. We evaluate the frameworks on venture commercialization and heart disease datasets. The optimization-based classifiers exhibit strong out-of-sample performance, demonstrating robust generalization and limited overfitting, and outperform benchmark machine learning classifiers while preserving a transparent and interpretable threshold-based decision structure.