Publications
2021
1.
Cueto-López, Nahúm; García-Ordás, María Teresa; Vitelli-Storelli, Facundo; Fernández-Navarro, Pablo; Palazuelos, Camilo; Alaiz-Rodríguez, Rocío
Evaluation of feature selection techniques for breast cancer risk prediction Artículo de revista
En: International Journal of Environmental Research and Public Health, vol. 18, no 20, pp. 10670, 2021, (Publisher: MDPI).
Resumen | Enlaces | BibTeX | Etiquetas: Breast Cancer, feature selection, Risk Prediction Model, stability
@article{cueto-lopez_evaluation_2021,
title = {Evaluation of feature selection techniques for breast cancer risk prediction},
author = {Nahúm Cueto-López and María Teresa García-Ordás and Facundo Vitelli-Storelli and Pablo Fernández-Navarro and Camilo Palazuelos and Rocío Alaiz-Rodríguez},
url = {https://www.mdpi.com/1660-4601/18/20/10670},
year = {2021},
date = {2021-01-01},
journal = {International Journal of Environmental Research and Public Health},
volume = {18},
number = {20},
pages = {10670},
abstract = {This study evaluates feature selection techniques combined with machine learning classifiers to improve breast cancer risk prediction. Using data from the MCC-Spain study (919 cases, 946 controls) with environmental and genetic features, the goal is to identify key risk factors and assess the stability of feature selection methods. SVM-RFE achieved the best performance, with a Logistic Regression model using its top-47 ranked features obtaining an AUC of 0.616 (5.8% improvement). SVM-RFE and Random Forest were the most stable selection methods, but SVM-RFE outperformed Random Forest in predictive accuracy.},
note = {Publisher: MDPI},
keywords = {Breast Cancer, feature selection, Risk Prediction Model, stability},
pubstate = {published},
tppubtype = {article}
}
This study evaluates feature selection techniques combined with machine learning classifiers to improve breast cancer risk prediction. Using data from the MCC-Spain study (919 cases, 946 controls) with environmental and genetic features, the goal is to identify key risk factors and assess the stability of feature selection methods. SVM-RFE achieved the best performance, with a Logistic Regression model using its top-47 ranked features obtaining an AUC of 0.616 (5.8% improvement). SVM-RFE and Random Forest were the most stable selection methods, but SVM-RFE outperformed Random Forest in predictive accuracy.
2.
Velasco-Mata, Javier; González-Castro, Víctor; Fidalgo, Eduardo; Alegre, Enrique
Efficient detection of botnet traffic by features selection and decision trees Artículo de revista
En: IEEE Access, vol. 9, pp. 120567–120579, 2021, (Publisher: IEEE).
Resumen | Enlaces | BibTeX | Etiquetas: Botnet Detection, Cybersecurity, feature selection, machine learning, Network Traffic Analysis
@article{velasco-mata_efficient_2021,
title = {Efficient detection of botnet traffic by features selection and decision trees},
author = {Javier Velasco-Mata and Víctor González-Castro and Eduardo Fidalgo and Enrique Alegre},
url = {https://ieeexplore.ieee.org/abstract/document/9523853},
year = {2021},
date = {2021-01-01},
journal = {IEEE Access},
volume = {9},
pages = {120567–120579},
abstract = {Botnets pose a major online threat, causing significant economic losses. With the rise of connected devices, analyzing large network traffic data is crucial. This study enhances botnet traffic classification by selecting the most relevant features using Information Gain and Gini Importance. Three feature subsets (5, 6, and 7 features) were tested with Decision Tree, Random Forest, and k-Nearest Neighbors on two datasets derived from CTU-13 (QB-CTU13 and EQB-CTU13). Results show that Decision Trees with a five-feature set achieved the best performance, with an 85% F1 score and an average classification time of 0.78 microseconds per sample.},
note = {Publisher: IEEE},
keywords = {Botnet Detection, Cybersecurity, feature selection, machine learning, Network Traffic Analysis},
pubstate = {published},
tppubtype = {article}
}
Botnets pose a major online threat, causing significant economic losses. With the rise of connected devices, analyzing large network traffic data is crucial. This study enhances botnet traffic classification by selecting the most relevant features using Information Gain and Gini Importance. Three feature subsets (5, 6, and 7 features) were tested with Decision Tree, Random Forest, and k-Nearest Neighbors on two datasets derived from CTU-13 (QB-CTU13 and EQB-CTU13). Results show that Decision Trees with a five-feature set achieved the best performance, with an 85% F1 score and an average classification time of 0.78 microseconds per sample.
2020
3.
Alaiz-Rodríguez, Rocío; Parnell, Andrew
A machine learning approach for lamb meat quality assessment using FTIR spectra Artículo de revista
En: Ieee Access, vol. 8, pp. 52385–52394, 2020, (Publisher: IEEE).
Resumen | Enlaces | BibTeX | Etiquetas: feature selection, food authenticity, neural networks, spectroscopy
@article{alaiz-rodriguez_machine_2020,
title = {A machine learning approach for lamb meat quality assessment using FTIR spectra},
author = {Rocío Alaiz-Rodríguez and Andrew Parnell},
url = {https://ieeexplore.ieee.org/abstract/document/9000861},
year = {2020},
date = {2020-01-01},
urldate = {2020-01-01},
journal = {Ieee Access},
volume = {8},
pages = {52385–52394},
abstract = {The food industry requires automatic methods to establish authenticity of food products. In this work, we address the problem of the certification of suckling lamb meat with respect to the rearing system. We evaluate the performance of neural network classifiers as well as different dimensionality reduction techniques, with the aim of categorizing lamb fat by means of spectroscopy and analyzing the features with more discrimination power. Assessing the stability of feature ranking algorithms also becomes particularly important. We assess six feature selection techniques (χ2, Information Gain, Gain Ratio, Relief and two embedded techniques based on the decision rule 1R and SVM (Support Vector Machine). Additionally, we compare them with common approaches in the chemometrics field like the Partial Least Square (PLS) model and Principal Component Analysis (PCA) regression. Experimental results with a fat sample dataset collected from carcasses of suckling lambs show that performing feature selection contributes to classification performance increasing accuracy from 89.70% with the full feature set to 91.80% and 93.89% with the SVM approach and PCA, respectively. Moreover, the neural classifiers yield a significant increase in the accuracy with respect to the PLS model (85.60% accuracy). It is noteworthy that unlike PCA or PLS, the feature selection techniques that select relevant wavelengths allow the user to identify the regions in the spectrum with the most discriminant power, which makes the understanding of this process easier for veterinary experts. The robustness of the feature selection methods is assessed via a visual approach.},
note = {Publisher: IEEE},
keywords = {feature selection, food authenticity, neural networks, spectroscopy},
pubstate = {published},
tppubtype = {article}
}
The food industry requires automatic methods to establish authenticity of food products. In this work, we address the problem of the certification of suckling lamb meat with respect to the rearing system. We evaluate the performance of neural network classifiers as well as different dimensionality reduction techniques, with the aim of categorizing lamb fat by means of spectroscopy and analyzing the features with more discrimination power. Assessing the stability of feature ranking algorithms also becomes particularly important. We assess six feature selection techniques (χ2, Information Gain, Gain Ratio, Relief and two embedded techniques based on the decision rule 1R and SVM (Support Vector Machine). Additionally, we compare them with common approaches in the chemometrics field like the Partial Least Square (PLS) model and Principal Component Analysis (PCA) regression. Experimental results with a fat sample dataset collected from carcasses of suckling lambs show that performing feature selection contributes to classification performance increasing accuracy from 89.70% with the full feature set to 91.80% and 93.89% with the SVM approach and PCA, respectively. Moreover, the neural classifiers yield a significant increase in the accuracy with respect to the PLS model (85.60% accuracy). It is noteworthy that unlike PCA or PLS, the feature selection techniques that select relevant wavelengths allow the user to identify the regions in the spectrum with the most discriminant power, which makes the understanding of this process easier for veterinary experts. The robustness of the feature selection methods is assessed via a visual approach.
4.
Alaiz-Rodríguez, Rocío; Parnell, Andrew
An information theoretic approach to quantify the stability of feature selection and ranking algorithms Artículo de revista
En: Knowledge-Based Systems, vol. 195, pp. 105745, 2020, (Publisher: Elsevier).
Resumen | Enlaces | BibTeX | Etiquetas: feature ranking, feature selection, Jensen-Shannon divergence, robustness, stability
@article{alaiz-rodriguez_information_2020,
title = {An information theoretic approach to quantify the stability of feature selection and ranking algorithms},
author = {Rocío Alaiz-Rodríguez and Andrew Parnell},
url = {https://www.sciencedirect.com/science/article/pii/S0950705120301593},
year = {2020},
date = {2020-01-01},
urldate = {2020-01-01},
journal = {Knowledge-Based Systems},
volume = {195},
pages = {105745},
abstract = {Feature selection is essential for high-dimensional data, but instability in algorithm outcomes can lead to varying feature rankings. This paper proposes using Jensen–Shannon divergence to measure the stability of feature selection methods, applicable to full and partial ranked lists. The approach emphasizes disagreements at the top of the list and offers desirable properties such as correction for change. The method is validated through experiments, including a food quality assessment, showing its effectiveness over traditional metrics like Spearman’s rank correlation.},
note = {Publisher: Elsevier},
keywords = {feature ranking, feature selection, Jensen-Shannon divergence, robustness, stability},
pubstate = {published},
tppubtype = {article}
}
Feature selection is essential for high-dimensional data, but instability in algorithm outcomes can lead to varying feature rankings. This paper proposes using Jensen–Shannon divergence to measure the stability of feature selection methods, applicable to full and partial ranked lists. The approach emphasizes disagreements at the top of the list and offers desirable properties such as correction for change. The method is validated through experiments, including a food quality assessment, showing its effectiveness over traditional metrics like Spearman’s rank correlation.
2019
5.
Alaiz-Rodríguez, Rocío; Cueto-López, Nahúm; García-Ordás, Maria Teresa; Dávila-Batista, Verónica; Moreno, Víctor; Aragonés, Nuria
A comparative study on feature selection for a risk prediction model for colorectal cancer Artículo de revista
En: Computer methods and programs in biomedicine, vol. 177, pp. 219–229, 2019, (Publisher: Elsevier).
Resumen | Enlaces | BibTeX | Etiquetas: algorithm stability, colorectal cancer, feature selection, ranking methods, risk prediction models
@article{alaiz-rodriguez_comparative_2019,
title = {A comparative study on feature selection for a risk prediction model for colorectal cancer},
author = {Rocío Alaiz-Rodríguez and Nahúm Cueto-López and Maria Teresa García-Ordás and Verónica Dávila-Batista and Víctor Moreno and Nuria Aragonés},
url = {https://arxiv.org/abs/2402.05293},
year = {2019},
date = {2019-01-01},
journal = {Computer methods and programs in biomedicine},
volume = {177},
pages = {219–229},
abstract = {The aim of this study is to evaluate risk prediction models to identify individuals at higher risk of developing colorectal cancer, focusing on feature selection methods. This is crucial for improving model performance, avoiding overfitting, and highlighting key risk factors. Additionally, the stability of feature selection/ranking methods is analyzed using conventional metrics and a visual approach proposed in this study.},
note = {Publisher: Elsevier},
keywords = {algorithm stability, colorectal cancer, feature selection, ranking methods, risk prediction models},
pubstate = {published},
tppubtype = {article}
}
The aim of this study is to evaluate risk prediction models to identify individuals at higher risk of developing colorectal cancer, focusing on feature selection methods. This is crucial for improving model performance, avoiding overfitting, and highlighting key risk factors. Additionally, the stability of feature selection/ranking methods is analyzed using conventional metrics and a visual approach proposed in this study.
2018
6.
Cueto-López, Nahúm; Alaiz-Rodríguez, Rocío; García-Ordás, María Teresa; González-Donquiles, Carmen; Martín, Vicente
Assessing feature selection techniques for a colorectal cancer prediction model Artículo de revista
En: International Joint Conference SOCO’17-CISIS’17-ICEUTE’17 León, Spain, September 6–8, 2017, Proceeding 12, pp. 471–481, 2018, (Publisher: Springer International Publishing).
Resumen | Enlaces | BibTeX | Etiquetas: colorectal cancer, feature selection, healthcare analytics, machine learning, risk prediction
@article{cueto-lopez_assessing_2018,
title = {Assessing feature selection techniques for a colorectal cancer prediction model},
author = {Nahúm Cueto-López and Rocío Alaiz-Rodríguez and María Teresa García-Ordás and Carmen González-Donquiles and Vicente Martín},
url = {https://link.springer.com/chapter/10.1007/978-3-319-67180-2_46},
year = {2018},
date = {2018-01-01},
journal = {International Joint Conference SOCO’17-CISIS’17-ICEUTE’17 León, Spain, September 6–8, 2017, Proceeding 12},
pages = {471–481},
abstract = {Risk prediction models for colorectal cancer help identify high-risk individuals and key risk factors. This study evaluates feature ranking algorithms in terms of stability and performance. Results show that Random Forest (RF) is the most stable but not the best-performing model, while SVM-wrapper and Pearson correlation achieve a balance between stability and predictive accuracy.},
note = {Publisher: Springer International Publishing},
keywords = {colorectal cancer, feature selection, healthcare analytics, machine learning, risk prediction},
pubstate = {published},
tppubtype = {article}
}
Risk prediction models for colorectal cancer help identify high-risk individuals and key risk factors. This study evaluates feature ranking algorithms in terms of stability and performance. Results show that Random Forest (RF) is the most stable but not the best-performing model, while SVM-wrapper and Pearson correlation achieve a balance between stability and predictive accuracy.
2017
7.
Fidalgo, Eduardo
Selection of relevant information to improve Image Classification using Bag of Visual Words Artículo de revista
En: ELCVIA: electronic letters on computer vision and image analysis, vol. 16, no 2, pp. 5–8, 2017.
Resumen | Enlaces | BibTeX | Etiquetas: Computer vision, feature selection, Fine-grained Recognition, Image classification
@article{fidalgo_selection_2017,
title = {Selection of relevant information to improve Image Classification using Bag of Visual Words},
author = {Eduardo Fidalgo},
url = {https://recercat.cat/handle/2072/428567},
year = {2017},
date = {2017-01-01},
journal = {ELCVIA: electronic letters on computer vision and image analysis},
volume = {16},
number = {2},
pages = {5–8},
abstract = {This PhD thesis addresses one of the main challenges in computer vision: image classification. With the rapid growth in the number of images, reliable classification has become increasingly important. The conventional classification pipeline involves extracting local image features, encoding them into a feature vector, and classifying them using a pre-trained model. The Bag of Words model and its extensions, such as pyramid matching and weighted schemes, have shown strong results. However, errors can occur in any step of the process, which may lead to a decline in classification performance. These errors can stem from multiple objects in an image, thin or small objects, incorrect annotations, or fine-grained recognition tasks. The thesis demonstrates that selecting high-quality features can significantly improve fine-grained classification, showing that a large training dataset is not always necessary to achieve good results.},
keywords = {Computer vision, feature selection, Fine-grained Recognition, Image classification},
pubstate = {published},
tppubtype = {article}
}
This PhD thesis addresses one of the main challenges in computer vision: image classification. With the rapid growth in the number of images, reliable classification has become increasingly important. The conventional classification pipeline involves extracting local image features, encoding them into a feature vector, and classifying them using a pre-trained model. The Bag of Words model and its extensions, such as pyramid matching and weighted schemes, have shown strong results. However, errors can occur in any step of the process, which may lead to a decline in classification performance. These errors can stem from multiple objects in an image, thin or small objects, incorrect annotations, or fine-grained recognition tasks. The thesis demonstrates that selecting high-quality features can significantly improve fine-grained classification, showing that a large training dataset is not always necessary to achieve good results.
8.
Fidalgo, Eduardo
Selection of relevant information to improve Image Classification using Bag of Visual Words Artículo de revista
En: 2017.
Resumen | Enlaces | BibTeX | Etiquetas: Computer vision, feature selection, Fine-grained Recognition, Image classification
@article{fidalgo_selection_2017-2,
title = {Selection of relevant information to improve Image Classification using Bag of Visual Words},
author = {Eduardo Fidalgo},
url = {https://www.raco.cat/index.php/ELCVIA/article/view/v16-n2-fidalgo},
year = {2017},
date = {2017-01-01},
abstract = {This PhD thesis tackles a major challenge in computer vision: image classification. With the rapid increase in the number of images, it is essential to classify them accurately and efficiently. The typical classification pipeline involves extracting image features, encoding them into vectors, and classifying them with a pre-trained model. The Bag of Words model and its variants, such as pyramid matching and weighted schemes, have proven to be effective. However, errors can occur at any stage, causing performance issues, especially when dealing with multiple objects, small or thin items, incorrect annotations, or fine-grained recognition. The thesis highlights the importance of good feature selection to enhance classification performance, showing that high-quality features can lead to improved fine-grained classification without requiring extensive training datasets.},
keywords = {Computer vision, feature selection, Fine-grained Recognition, Image classification},
pubstate = {published},
tppubtype = {article}
}
This PhD thesis tackles a major challenge in computer vision: image classification. With the rapid increase in the number of images, it is essential to classify them accurately and efficiently. The typical classification pipeline involves extracting image features, encoding them into vectors, and classifying them with a pre-trained model. The Bag of Words model and its variants, such as pyramid matching and weighted schemes, have proven to be effective. However, errors can occur at any stage, causing performance issues, especially when dealing with multiple objects, small or thin items, incorrect annotations, or fine-grained recognition. The thesis highlights the importance of good feature selection to enhance classification performance, showing that high-quality features can lead to improved fine-grained classification without requiring extensive training datasets.
2011
9.
Guzmán-Martínez, Roberto; Alaiz-Rodríguez, Rocío
Feature selection stability assessment based on the jensen-shannon divergence Artículo de revista
En: Joint European conference on machine learning and knowledge discovery in databases, pp. 597–612, 2011, (Publisher: Springer Berlin Heidelberg Berlin, Heidelberg).
Resumen | Enlaces | BibTeX | Etiquetas: feature ranking, feature selection, Jensen-Shannon divergence, robustness, stability
@article{guzman-martinez_feature_2011,
title = {Feature selection stability assessment based on the jensen-shannon divergence},
author = {Roberto Guzmán-Martínez and Rocío Alaiz-Rodríguez},
url = {https://link.springer.com/chapter/10.1007/978-3-642-23780-5_48},
year = {2011},
date = {2011-01-01},
journal = {Joint European conference on machine learning and knowledge discovery in databases},
pages = {597–612},
abstract = {This work introduces a generalized, information-theoretic metric based on Jensen-Shannon divergence to evaluate the stability of feature selection and ranking techniques. Unlike traditional metrics, it can handle full rankings, top-k lists, and partial rankings while giving more weight to top-ranked features. The method is compared with existing metrics like Spearman correlation and Kuncheva’s index through synthetic and real datasets.},
note = {Publisher: Springer Berlin Heidelberg Berlin, Heidelberg},
keywords = {feature ranking, feature selection, Jensen-Shannon divergence, robustness, stability},
pubstate = {published},
tppubtype = {article}
}
This work introduces a generalized, information-theoretic metric based on Jensen-Shannon divergence to evaluate the stability of feature selection and ranking techniques. Unlike traditional metrics, it can handle full rankings, top-k lists, and partial rankings while giving more weight to top-ranked features. The method is compared with existing metrics like Spearman correlation and Kuncheva’s index through synthetic and real datasets.