Hugo De Oliveira

A Python framework to simulate a debate between two ChatGPT agents. The script will generate the discussion by providing a subject for the debate and an opinion as well as a tone for the argumentation of the two agents.

A Python project to create a dashboard automating analytics with ChatGPT. After loading CSV data, use natural language to ask ChatGPT to plot a chart. The chart will be displayed in the dashboard, allowing for export. The module allows for context, making possible iterations to refine the visualization.

This project focused on developing an end-to-end methodology to predict a pathway-related outcome and identifying predictive factors using autoencoders. The method was tested in a case study, predicting short-term mortality after the implementation of an implantable cardioverter-defibrillator.

The “Bow-tie” optimal pathway discovery analysis uses large clinical event datasets to map clinical pathways and to visualize risks (improvement opportunities) before and outcomes, after a specific clinical event. This proof-of-concept study assesses the use of NHS Hospital Episode Statistics (HES) in England as a potential clinical event dataset for this pathway discovery analysis approach.

Process mining is a suitable method for knowledge extraction from patient pathways. Structured in event logs, medical events are complex, often described using various medical codes. Finding an efficient method of labeling these events before applying process mining analysis was challenging.

This project focused on developing an innovative methodology to handle the complexity of events in medical event logs. Based on auto-encoding, accurate labels are created by clustering similar events in latent space. Moreover, the explanation of created labels is provided by the decoding of the corresponding events.

This project focuses on solving the problem of determining the optimal process model of an event log of traces of events with temporal information. We introduced a new formalism, along with a Tabu search algorithm to determine the optimal process model that maximizes the traces' representation subject to the constraints of the maximal number of nodes and arcs. We then conducted a healthcare case study to demonstrate the applicability of the approach for clinical pathway modeling. Special attention was paid to readability, so those final users could interpret the process mining results.

In this project, a benchmark of seven machine learning algorithms was performed on binary classification tasks of hospital data. We then tested seven algorithms on three data sets extracted from the French national hospital database. Lastly, we used an efficient global optimization algorithm to solve the hyperparameter tuning problem.

This project focuses on the study of treatment sequences, particularly the extraction of patterns from nonclinical claim databases through clustering. For this purpose, the TAK algorithm was proposed and demonstrated its usefulness. However, the scalability of the TAK algorithm regarding the number of patients was an issue; the method was impossible to use in practice for thousands of patients. For this purpose, we developed an extension of the TAK algorithm. Referred to as Meta-TAK, this method appears to be robust and computationally efficient.