Solving Feature Selection Problems with Quantum Algorithms on Real Financial Data

• Autor(a) principal: Galvão, Lucas
• Data de Publicação: 2024
• Outros Autores: Pires, Otto, Chagas, Yan Alef, Fraga, Maria Heloísa, Moret, Marcelo A.
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Vetor (Online)
• Texto Completo: https://periodicos.furg.br/vetor/article/view/18358
• topic: Feature selection; Financial Data; Quantum Annealing; QUBO; Machine Learning; Computational modeling; Machine Learning; Seleção de características; Dados financeiros; Quantum Annealing; QUBO; Aprendizado de Máquina; Modelagem computacional; Aprendizado de máquina

The financial sector faces significant challenges when dealing with high-dimensional datasets and a limited number of samples, making it difficult to build robust predictive models. Traditional machine learning techniques help mitigate these problems, but the presence of irrelevant and redundant features increases computational complexity. This article presents the application of quantum algorithms in feature selection using real data from the financial sector, demonstrating that these algorithms can improve the efficiency and accuracy of predictive models. The approach involves formulating the problem in terms of Unconstrained Quadratic Binary Optimization (QUBO), and its solution implemented in quantum annealer simulators. The experiments show promising results, which are analyzed using the Akaike Information Criterion metric. The results suggest that variational quantum algorithms have great application potential compared to traditional feature selection techniques.