258178 Network-Induced Supervised Learning: Balancing Interpretation and Prediction Ability in Classification and Regression Tasks

Monday, October 29, 2012
Hall B (Convention Center )
Marco Reis, Department of Chemical Engineering, University of Coimbra, Coimbra, Portugal

Current supervised learning approaches are strongly focused on optimizing estimation accuracy metrics, leaving to a secondary concern the interpretation of the results produced (Hastie et al., 2001; Reis & Saraiva, 2004, 2005, 2010). However, in the analysis of complex systems, one of the main interests is precisely the induction of relevant associations, in order to understand or clarify the way systems operate. The importance of this issue is not new and has motivated the development of approaches that try to improve the interpretation of results, while maintaining the quality of predictions, such as O-PLS (Trygg & Wold, 2002).

In this communication, we present two new frameworks for addressing supervised learning problems (classification and regression), that incorporate interpretational-oriented analysis features right from the onset of the analysis. These features constrain the predictive space, in order to introduce interpretable elements in the final model. After extensive testing, we verified that such constraints do not usually compromise the methods’ performance, when compared to their unconstrained versions, and sometimes even led to improvements in prediction ability, due to the use of more parsimonious and robust models.

The frameworks, called Network-Induced Classification (NI-C) and Network-Induced Regression (NI-R), share a common methodological backbone, based on a variable clustering algorithm that takes into consideration the local associations among variables, in order to identify meaningful clusters of functionally related variables. These groups of variables constitute the basic predictive atoms that would be selectively requested at a second stage, for forming the predictive models. In the proposed communication, the methods will be described in detail, and results will be presented for several real world data sets.


Hastie, T., Tibshirani, R. & Friedman, J. (2001). The elements of statistical learning.  NY: Springer.

Reis, M. S. & Saraiva, P. M. (2004). A comparative study of linear regression methods in noisy environments. Journal of Chemometrics, 18(12), 526-536.

Reis, M. S. & Saraiva, P. M. (2005). Integration of data uncertainty in linear regression and process optimization. AIChE Journal, 51(11), 3007-3019.

Reis, M. S. & Saraiva, P. M. (2010). Analysis and classification of the paper surface. Industrial & Engineering Chemistry Research, 49(5), 2493-2502.

Trygg, J. & Wold, S. (2002). Orthogonal projections to latent structures (o-pls). Journal of Chemometrics, 16(3), 119-128.

Extended Abstract: File Not Uploaded