Classification is one of the central tasks in machine learning: given a set of observations with known class labels, learn a model that can correctly classify new, unseen observations. My research on classification methods has developed novel classifiers based on probabilistic graphical models, ensemble methods, and evolutionary algorithms, with applications to remote sensing, energy systems, building management, and neuroscience.

Probabilistic Classifiers

Bayesian Network Classifiers
Development and evaluation of Bayesian network classifiers, from the simple naïve Bayes model to more expressive general Bayesian network classifiers. Research on how the structure of the Bayesian network affects classification accuracy, and on learning algorithms that optimize the structure specifically for classification rather than density estimation.
Vine Copula Classification
Application of vine copula models to classification tasks. Vine copulas provide a flexible framework for modeling the joint distribution of continuous features, allowing classifiers to capture complex multivariate dependency structures without making strong distributional assumptions.
Searchlight Classification for Brain Decoding
Application of classification methods to brain decoding, using searchlight analysis to identify which brain regions carry information about specific stimuli or cognitive states. Development of domain adaptation methods to transfer classification models across different experimental paradigms (e.g., visual perception to mental imagery).

Ensemble Methods

Random Vector Functional Link Forests
Research on random vector functional link (RVFL) forests and extreme learning forests applied to UAV automatic target recognition. RVFL networks are single-hidden-layer networks with randomly initialized and fixed input weights; forests of RVFL networks provide competitive performance with much lower training cost than deep learning alternatives.
Stacking LLM Predictions for Classification
Stacking (ensemble learning) applied to the predictions of multiple large language models for anomaly classification. By combining the predictions of diverse LLM models through a meta-learner, accuracy and explainability are improved compared to using any individual model alone.

Feature Selection

Filter-Based Feature Selection for Multi-target Regression
Development of filter method-based feature selection for multi-target regression problems with unattributed-identity data (where the correspondence between input and output samples is partially unknown). Research on how different filter measures handle the multi-target setting and the challenges posed by identity ambiguity.
Feature Selection for Anomaly Classification
Feature selection methods for anomaly classification in building management systems. Research on how to identify the most relevant features for classifying faults and anomalies in fan coil units and other building components, with an emphasis on causality and explainability of the selected features.

Time-Series Classification

Classification of Household Devices from Energy Time Series
Classification of household devices from electricity usage time-series data (non-intrusive load monitoring). Research on the impact of different imputation methods on classification accuracy and on the features that best characterize different device types from their electricity consumption patterns.
Fault Classification in Building Systems
Classification of faults in fan coil units from multivariate time-series sensor data. Research on the impact of temporal features and the choice of time-series representation on fault classification accuracy. Development of methods that exploit temporal dependencies in the data for more accurate fault detection and diagnosis.

Applications

Sand Dune Detection on Mars
Detection of sand dunes on Mars using a vine copula-based classification approach applied to orbital imagery. This application demonstrates the versatility of copula-based classifiers for remote sensing problems where the distribution of image features is non-Gaussian and exhibits complex multivariate dependencies.
Award Price Estimation for Public Procurement
Application of machine learning classification and regression algorithms to estimate the award price in public procurement auctions. Case study with tender data from Spain, examining how well different machine learning models can predict procurement outcomes from available features of the tender.

Selected Publications