A carregar...
Publicação
DeepPAY: A Framework for Dimensionality Reduction and Interactive Exploration
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.53 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
As payments become more digitalized and interconnected, the complexity and volume of the
data they generate continue to grow, resulting in new challenges for analysis and
interpretation. These challenges are critical in contexts where timely insight and transparency
are essential. This study introduces a visual analytics framework designed to explore highdimensional payment data through a combination of dimensionality reduction (PCA and
UMAP), clustering techniques, and an autoencoder–isolation forest model for anomaly
detection. The solution is implemented as a modular Dash application that supports dynamic
interaction, enabling users to uncover latent structures, identify behavioural clusters, and
detect anomalous or inconsistent records. Two usage scenarios are explored, using payments
data collected by Banco de Portugal: one that examines the clustering patterns and network
topology of interbank transactions, and another that focuses on detecting abrupt changes and
irregularities in paymentseries. Together, these use cases demonstrate the system’s potential
to support both structural and temporal analyses. Although the framework was developed
with a focus on payment systems, the approach is sufficiently general to be applied to other
domains involving high-dimensional transactional information, such as stock exchange
operations or insurance records. By bringing together analytical depth and interpretability,
this work contributes to the design of transparent and flexible tools for navigating complex
data ecosystems.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Data Science
Palavras-chave
anomaly detection clustering dash application high-dimensional data network analysis payment data