New Preprint: Functional-Ordinal Canonical Correlation Analysis (foCCA)

Many sensor-driven applications require more than simple label prediction. Instead, the aim is often to assess ordered states such as deterioration levels, quality grades, or risk stages from functional data, for example time-series or spectral profiles measured by optical sensors.

In a new preprint, researchers from Politecnico di Milano and Eberhard Karls Universität Tübingen introduce Functional-Ordinal Canonical Correlation Analysis (foCCA). The proposed method performs dimensionality reduction for functional data while explicitly incorporating the ordinal structure of the response variable.

foCCA is designed to maximize separation between consecutive ordinal levels and to capture relative dissimilarities between neighboring stages. By doing so, it improves both prediction performance and interpretability compared to existing approaches.

Simulation studies show that foCCA outperforms current state-of-the-art methods in a reduced feature space. In addition, a real-world case study on predicting antigen concentration levels from optical biosensor signals demonstrates the practical advantages of the proposed approach.

🔗 Read more: arxiv.org/pdf/2501.18317 https://www.tandfonline.com/doi/abs/10.1080/00401706.2025.2584479