Arinan Dourado, PhD

This project develops trustworthy and explainable machine-learning methods to support the preoperative characterization of meningiomas and prediction of postoperative outcomes. The research integrates clinical, radiographic, laboratory, demographic, and disease-specific information to produce patient-level predictions accompanied by uncertainty estimates and interpretable explanations. The broader objective is to help clinicians understand not only what a model predicts, but why the prediction was produced, when it should be trusted, and how alternative patient conditions could affect the result.

The TRUST (Trustworthy, Resilient, Uncertainty-Aware Systems and Technologies) Research Initiative focuses on how trust can be engineered into intelligent systems. This research effort aims to develop a multidimensional framework for evaluating AI through predictive robustness, domain consistency, operational familiarity, and reproducibility. The objective is to produce actionable assurance evidence for high-consequence AI-enabled systems.

This research applies explainable machine learning to personalized decision-making in total knee arthroplasty. Current studies investigate procedure selection, postoperative outcomes, and the identification of patients likely to achieve high patient-reported outcome scores. By integrating demographic characteristics, preoperative outcomes, intraoperative alignment measurements, and follow-up data, the project seeks to produce reliable predictions with patient-specific explanations that can support clinical judgment and shared decision-making.

This project investigates how explainable machine learning can help universities better understand and support student persistence in engineering. Neural networks, random forests, and Bayesian learning methods are used to identify patterns associated with continued enrollment and degree progression, while explanation methods examine which institutional, academic, and potentially malleable factors influence individual predictions. The goal is to develop responsible educational analytics that inform timely, evidence-based mentoring and student support without treating predictions as deterministic judgments.