University Work
This section includes my Final Degree Project (TFG), Final Master’s Project (TFM), and Collective TFM, where I have applied advanced methodologies in Statistics, Machine Learning, and Deep Learning to real-world problems.
Optimization of Accuracy and Efficiency in Real-Time Emotion Recognition Using Convolutional Neural Networks in Video Surveillance Systems
Artificial Intelligence has established itself as one of the most dynamic and rapidly evolving fields in recent years, enabling the development of increasingly precise tools and techniques for analyzing complex data. In particular, Neural Networks have become a highly effective methodology for recognizing patterns in complex data, including facial expressions.
This Master’s Final Project, awarded Honors and an outstanding grade of 9.6, was meticulously developed and presented before a panel of experts. Its primary objective is to create a tool based on Convolutional Neural Networks (CNNs) for identifying the four standard emotions that individuals can express exclusively through facial expressions: joy, disgust, surprise, and neutral. Through this model, individuals can be accurately classified into different emotional states, enabling the interpretation of both positive and negative experiences.
The model developed in this project serves as the minimum viable product (MVP) for a startup aiming to integrate Artificial Intelligence into everyday life. The first initiative of this company is to implement the model in video surveillance cameras to analyze customer emotions in various establishments. This innovative approach holds significant value for businesses, as it allows business owners and managers to enhance the customer experience and satisfaction.
Ultimately, this Master’s Final Project contributes to the advancement of Artificial Intelligence technologies for emotion recognition and explores their potential impact on enhancing user experience across various industries. Furthermore, this research represents a first step in establishing an innovative and disruptive AI-driven company, following an end-to-end methodology—from image acquisition to model deployment.
Development of a Neural Network Model for Detecting Environmental Pollution Levels in the U.S.
This summary highlights the achievements of Team PFG 7 from the University of Navarra in their collective master’s final project, which was awarded a grade of 8.5. The project focused on the field of data science, aiming to predict which U.S. states will fail to meet air quality standards set by the Environmental Protection Agency (EPA).
The team worked with a large dataset containing 2 million records and 55 variables, collecting relevant information to assess air quality. Through a detailed problem analysis, the team conducted data processing and treatment, identifying the most significant variables and defining a target variable to classify air quality levels.
Several models were explored, including Random Forest, but ultimately, a Neural Network model was implemented, proving to be the most effective for the classification task. The results were highly promising—not only was the model able to predict the most dominant air quality category, but the team also successfully addressed bias in the dataset, achieving accurate predictions across all categories.
The collective master’s final project of Team PFG 7 was a success, standing out for its rigorous data analysis, model selection, and innovative approach to air quality prediction. These findings have the potential to make a significant impact on environmental protection efforts and aid in data-driven decision-making to improve air quality in the United States.
Decision Support Systems for the Design of a Hospital Emergency Service
Below is a summary of the research project I conducted as part of my Final Degree Project (TFG), which was awarded a perfect score of 10 by a panel of experts. Additionally, this project was presented at the 6th Scientific Congress of Young Researchers in Experimental Design and Data Science, held in Pamplona on June 5th, 6th, and 7th, 2023.
The COVID-19 crisis had a direct negative impact on healthcare management, significantly increasing waiting timesand overcrowding in hospital emergency services, both public and private.
To address this issue, I developed a mathematical algorithm in R capable of simulating hospital emergency service operations. The project resulted in a decision-support tool that analyzes patient service times based on various hospital-specific parameters. This tool enables the optimization of emergency department design and resource allocation, improving efficiency and patient flow management.
The interpretation of simulation results is based on queueing models, specifically queueing theory. Therefore, it was essential to establish a clear understanding of the theoretical foundations presented in this project. The performance measures of queueing theory, along with patient flow tracking within the system, are visualized through a basic interface that allows users to query results and draw meaningful conclusions.