Quantum Implementation of Quadratic Programming for Feature Selection

This work explores the quantum implementation of Quadratic Programming for Feature Selection (QPFS), a method designed to address the challenges of high-dimensional datasets in machine learning. Feature selection plays a critical role in improving model performance, reducing computational complexity, and mitigating overfitting. QPFS formulates feature selection as a quadratic optimization problem that balances feature relevance and redundancy. We investigate the scalability challenges of QPFS and apply the Nystrom approximation to reduce computational cost. To demonstrate the feasibility of quantum acceleration, the QPFS algorithm was implemented on QCI’s Dirac-3 quantum platform, which leverages high-dimensional qudits for solving optimization problems. Experiments were conducted on the Taiwanese Bankruptcy Prediction dataset, with alternative datasets identified for future exploration. The results highlight key differences between classical and quantum implementations: while the classical solver produced rank-based weights for features, the Dirac-3 quantum solver effectively suppressed redundant and irrelevant features by assigning them near-zero values. This property is significant in simplifying feature selection without additional manual thresholding. The project provides insights into the integration of quantum optimization methods with machine learning pipelines, showing promising directions for scalable feature selection in finance, healthcare, and other high-dimensional domains. The full implementation, including both classical and quantum code, is available on GitHub: https://github.com/Hope-Alemayehu/QPFS-using-Dirac3.