32. Beyond Traditional PCA: The Two-Step-SDP Algorithm for Data Analysis
Invited abstract in session WC-2: Conic and Semidefinite Optimization, stream Conic optimization: theory, algorithms and applications.
Wednesday, 10:05 - 11:20Room: M:O
Authors (first author is the speaker)
| 1. | Eloisa Macedo
|
| Department of Mechanical Engineering, University of Aveiro |
Abstract
Analysing real-world data is fundamental for informed decision-making in various fields, from business and science, to healthcare and government. Due to ever-growing advances in sensing data, e.g., in the context of smart cities, large databases of objects and attributes can be difficult to analyse and extract meaningful information. Many data analysis frameworks for extracting hidden patterns in data have been proposed in the literature. Some methodologies allow to reduce the dimension of both objects and variables. In this context, the Two-Step-SDP methodology is relevant due to its light approach that incorporates a clustering on both objects and attributes and reveals good computational performance when compared to similar methodologies. First, the Two-Step-SDP algorithm focuses on solving two relaxed Semidefinite Programming (SDP) clustering models. These solutions provide starting points (centroids) for the clustering process of both sets of objects and attributes. Then, the algorithm iteratively refines these initial centroids to find optimal clusters, with the main objective of maximizing the between cluster deviance in a lower-dimensional space. The obtained solutions allow to unveil the underlying structure of the data through nonoverlapping clusters, which is important to ease data interpretation. The algorithm is applied to real road traffic-related data and unlocks the true potential of extracting information behind data.
Keywords
- Conic and semidefinite optimization
Status: accepted
Back to the list of papers