DSpace Repository :: Browsing by Author "Maldonado, Sebastián"

Browsing by Author "Maldonado, Sebastián"

Now showing 1 - 20 of 22

33
(33, 33) Vairetti, Carla; Assadi, José Luis; Maldonado, Sebastián; 33
Imbalanced classification is a well-known challenge faced by many real-world applications. This issue occurs when the distribution of the target variable is skewed, leading to a prediction bias toward the majority class. With the arrival of the Big Data era, there is a pressing need for efficient solutions to solve this problem. In this work, we present a novel resampling method called SMOTENN that combines intelligent undersampling and oversampling using a MapReduce framework. Both procedures are performed on the same pass over the data, conferring efficiency to the technique. The SMOTENN method is complemented with an efficient implementation of the neighborhoods related to the minority samples. Our experimental results show the virtues of this approach, outperforming alternative resampling techniques for small- and medium-sized datasets while achieving positive results on large datasets with reduced running times.
33
(33, 33) Maldonado, Sebastián; López, Julio; Vairetti, Carla; 33
In this paper, we propose three novel profit-driven strategies for churn prediction. Our proposals extend the ideas of the Minimax Probability Machine, a robust optimization approach for binary classification that maximizes sensitivity and specificity using a probabilistic setting. We adapt this method and other variants to maximize the profit of a retention campaign in the objective function, unlike most profit-based strategies that use profit metrics to choose between classifiers, and/or to define the optimal classification threshold given a probabilistic output. A first approach is developed as a learning machine that does not include a regularization term, and subsequently extended by including the LASSO and Tikhonov regularizers. Experiments on well-known churn prediction datasets show that our proposal leads to the largest profit in comparison with other binary classification techniques.
33
(33, 33) Maldonado, Sebastián; Vairetti, Carla; Fernandez, Alberto; Herrera, Francisco; 33
The Synthetic Minority Over-sampling Technique (SMOTE) is a well-known resampling strategy that has been successfully used for dealing with the class-imbalance problem, one of the most challenging pattern recognition tasks in the last two decades. In this work, we claim that SMOTE has an important issue when defining the neighborhood in order to create new minority samples: the use of the Euclidean distance may not be suitable in high-dimensional settings. Our hypothesis is that the use of a weighted metric that does not assume that all features are equally important could improve performance in the presence of noisy/redundant variables. In this line, we present a novel SMOTE-like method that uses the weighted Minkowski distance for defining the neighborhood for each example of the minority class. This methodology leads to a better definition of the neighborhood since it prioritizes those features that are more relevant for the classification task. A complementary advantage of the proposal is performing feature selection since attributes can be discarded when their corresponding weights are below a given threshold. Our experiments on 42 class-imbalance datasets show the virtues of the proposed SMOTE variant, achieving the best predictive performance when compared with the traditional SMOTE approach and other recent variants on low- and high-dimensional settings, handling issues such as class overlap and hubness adequately without increasing the complexity of the method.
33
(33, 33) Maldonado, Sebastián; Vairetti, Carla; Figueroa, Ignacio; 33
As with any task, the process of building machine learning models can benefit from prior experience. Meta-learning for classifier selection leverages knowledge about the characteristics of different datasets and/or the past performance of machine learning techniques to inform better decisions in the current modeling process. Traditional meta-learning approaches first collect metadata that describe this prior experience and then use it as input for an algorithm selection model. In this paper, however, a one-step scheme is proposed in which convolutional neural networks are trained directly on tabular datasets for binary classification. The aim is to learn the underlying structure of the data without the need to explicitly identify meta-features. Experiments with simulated datasets show that the proposed approach achieves near-perfect performance in identifying both linear and nonlinear patterns, outperforming the conventional two-step method based on meta-features. The method is further applied to real-world datasets, providing recommendations on the most suitable classifiers based on the data's inherent structure.
33
(33, 33) Carrasco, Miguel; Vairetti, Carla; López, Julio; Maldonado, Sebastián; 33
Kernel methods are crucial in machine learning due to their ability to model nonlinear relationships in data. Among these, Support Vector Machine (SVM) is widely recognized for its robust performance and appealing optimization properties. In this work, we build upon recent advancements in SVM variants to propose five novel models specifically designed for multiclass learning. In particular, we introduce One-vs-One and One-vs-All versions of the nonparallel hyperplane SVM and improved twin SVM, along with a unified optimization variant (all-together) of the former method for nonlinear multiclass classification. Our empirical evaluation, conducted on 11 datasets and 12 multiclass classifiers, shows the superiority of our methods: four out of the five proposed models rank among the top performers and consistently outperform alternative approaches in terms of balanced accuracy. Additionally, a statistical test was performed, showing significant differences among the classifiers.
33
(33, 33) Sánchez, Catalina; Maldonado, Sebastián; Vairetti, Carla; 33
Debt collection is a very important business application of predictive analytics. This task consists of foreseeing repayment chances of late payers. In this sense, contact centers have a central role in debt collection since it improves profitability by turning monetary losses into a direct benefit to banks and other financial institutions. In this paper, we study the influence of contact center variables in predictive models for debt collection, which are combined with the financial information of late payers. We explore five different variants of three predictive analytics tasks: (1) the probability of successfully contacting a late payer, (2) the probability of achieving a contact that results in a promise to pay a debt, and (3) the probability that a defaulter repays his/her arrears. Four research questions are developed in the context of debt collection analytics and empirically discussed using data from a Chilean financial institution. Our results show the positive impact of the combination of the two data sources in terms of predictive performance, confirming that valuable information on late payers can be collected from contact centers.
33
(33, 33) Latorre, Paolo; López-Ospina, Héctor; Maldonado, Sebastián; Guevara, C. Angelo; Pérez, Juan; 33
Employee turnover significantly impacts organizations, particularly those with substantial investments in training their workforce. To mitigate these effects, we propose a Prescriptive Human Resources Analytics approach that optimizes employee benefits to minimize total costs, focusing on turnover management The methodology models employee decision-making using a discrete choice model, with parameters estimated through maximum likelihood. We solve the resulting nonlinear optimization problem with a heuristic tailored to the problem's complexity. We applied this methodology to a hospital case study, which was used to enhance the transportation system as an employee benefit, considering the associated turnover costs. The results demonstrate that our approach can reduce total costs, optimize the usage level of the designed benefits, and increase employee satisfaction. This research provides a robust framework for data-driven decision-making in HR, offering practical tools for improving employee retention strategies.
33
(33, 33) Vairetti, Carla; Vargas, Raimundo; Sánchez, Catalina; García, Andrés; Armelini, Guillermo; Maldonado, Sebastián; 33
In this study, we present a novel cost-sensitive approach for uplift modeling in the context of cross-selling and workforce analytics. We leverage referrals from sales agents across business units to estimate the individual treatment effects of incentives on the cross-selling outcomes within a company. Uplift modeling is employed to predict relationships between salespeople that should be encouraged based on the probability of successful cross-selling - defined when a customer accepts the product suggested by sales agents. We conducted experiments on data from a Chilean financial group, evaluating both statistical and profit metrics. Exploring various machine learning classifiers for predictive purposes, we observed a significant improvement over the current approach, which exhibits an uplift below 0.01. Finally, we show that selecting the best classifier with profit metrics results in a 31.6% improvement in terms of average customer profit. This emphasizes the importance of defining an adequate compensation scheme and integrating it into the modeling process.
33
(33, 33) Aldunate, Ángeles; Maldonado, Sebastián; Vairetti, Carla; Armelini, Guillermo; 33
It is of utmost importance for marketing academics and service industry practitioners to understand the factors that influence customer satisfaction. This study proposes a novel framework to analyze open-ended survey data and extract drivers of customer satisfaction. This is done automatically via deep learning models for natural language processing. According to 11 drivers acknowledged by the marketing literature to determine customer experience, the data is cast into a multi-label classification problem. This expert system not only supports the automatic analysis of new data but also ranks the drivers according to their importance to various service industries and provides important insights into their applications. Experiments carried out using 25,943 customer survey responses related to 39 service companies in 13 different economic sectors show that the drivers can be identified accurately.
33
(33, 33) Maldonado, Sebastián; Vairetti, Carla; Jara, Katherine; Carrasco, Miguel; López, Julio; 33
In this paper, we propose a novel adaptive loss function for enhancing deep learning performance in classification tasks. Specifically, we redefine the cross-entropy loss to effectively address class-level noise conditions, including the challenging problem of class imbalance. Our approach introduces aggregation operators to improve classification accuracy. The rationale behind our proposed method lies in the iterative up-weighting of class-level components within the loss function, focusing on those with larger errors. To achieve this, we employ the ordered weighted average (OWA) operator and combine it with an adaptive scheme for gradient-based learning. The main finding is that our method outperforms other commonly used loss functions, such as the standard cross-entropy or focal loss, across various binary and multiclass classification tasks. Furthermore, we explore the influence of hyperparameters associated with the OWA operators and propose a default configuration that performs well across different experimental settings.
33
(33, 33) Biskupovic, Fernando; Rosenberg, Flavia; Searle, Luz María; Ramírez, Pamela; Larrañaga, María Jesús; Maldonado, Sebastián; Vairetti, Carla; Oyonarte, Rodrigo; 33
Background: Assessing a patient's maturation status is essential for treatment planning in dentofacial orthopedics. Dental development, as classified by Demirjian's method into eight stages, is a reliable indicator of skeletal maturity relative to the pubertal growth spurt. Automating this assessment may improve efficiency by reducing subjectivity and supporting timely orthodontic interventions. Methods: A cross-sectional study was conducted using segmented panoramic radiographs to classify the maturation stages of lower second and third molars. These classifications served as training data for machine learning models using four convolutional neural network (CNN) architectures: Xception, ResNet, MobileNet, and Inception. Model performance was evaluated on three datasets: second and third molars combined (ST), second molars only (S), and third molars only (T). Gradient-weighted Class Activation Mapping (Grad-CAM) was employed to visualize model attention. Results: A total of 1805 images were analyzed. Inception achieved the best performance in both the ST dataset (accuracy 0.96, precision 0.86, recall 0.85, and F1 score 0.85) and the S dataset (accuracy 0.98, precision 0.92, recall 0.91, and F1 score 0.89). For the T dataset, ResNet performed the best (accuracy 0.96, precision 0.94, recall 0.95, and F1 score 0.81). Inter-examiner agreement was high, with a mean kappa coefficient of 0.94. Grad-CAM heat maps confirmed that the model focused on relevant dental structures. Conclusions: The proposed deep learning system, especially the Inception model, demonstrated high accuracy and strong agreement with experts when classifying dental maturation stages. These findings support its use as a complementary diagnostic tool to aid clinical decision-making in growth assessment.
33
(33, 33) Maldonado, Sebastián; López, Julio; Vairetti, Carla; 33
The predictive performance of classification methods relies heavily on the nature of the environment, as in the joint distribution of inputs and outputs may evolve over time. This issue is known as dataset shift. Given that most statistical and machine learning techniques assume that the training sample is drawn from the same distribution as the test data used for evaluation, an appreciable amount of researchers and practitioners tend to ignore this issue at the model construction stage. In this paper, we propose a novel Fuzzy Support Vector Machine strategy, in which the traditional hinge loss function is redefined to account for dataset shift. Additionally, we propose a general version of this loss function applying aggregation operators in order to improve performance by dealing with dataset shift via fuzzy logic. Originally developed as linear approaches, our proposals are extended to kernel-based classification for non-linear machine learning. Our methods are able to perform best compared to traditional classifiers in terms of out-of-time prediction using simulated and real-world dataset for credit scoring, confirming the theoretical virtues of our approach.
33
(33, 33) García, Maximiliano; Maldonado, Sebastián; Vairetti, Carla; 33
In this paper, we present a novel approach for n-gram generation in text classification. The a-priori algorithm is adapted to prune word sequences by combining three feature selection techniques. Unlike the traditional two-step approach for text classification in which feature selection is performed after the n-gram construction process, our proposal performs an embedded feature elimination during the application of the a-priori algorithm. The proposed strategy reduces the number of branches to be explored, speeding up the process and making the construction of all the word sequences tractable. Our proposal has the additional advantage of constructing a low-dimensional dataset with only the features that are relevant for classification, that can be used directly without the need for a feature selection step. Experiments on text classification datasets for sentiment analysis demonstrate that our approach yields the best predictive performance when compared with other feature selection approaches, while also facilitating a better understanding of the words and phrases that explain a given task; in our case online reviews and ratings in various domains.
33
(33, 33) Vairetti, Carla; Gennaro, Franco; Maldonado, Sebastián; 33
In this paper, we propose a novel matching strategy to correct for confounding in uplift modeling. Our method, called propensity score oversampling and matching (ProSOM), extends the well-known propensity score matching (PSM) technique by addressing one of its main limitations: dealing with small datasets that face an imbalance in the distribution of the causal variable. Apart from this, we also face the additional complexity of dealing with class labels. The proposed method establishes a parallel between uplift modeling and class-imbalance classification as it extends existing oversampling techniques to create synthetic elements from the treatment group. We design an algorithm that performs classaware data oversampling in the treatment group, and then it matches samples from this group with the control group. This can be seen as a novel hybrid undersampling-oversampling solution for causal learning. Experiments on five datasets show the virtues of ProSOM in terms of predictive performance, achieving the best Qini coefficient for all five datasets in relation to PSM and other resampling solutions.
33
(33, 33) Vairetti, Carla; Martínez-Cámara, Eugenio; Maldonado, Sebastián; Luzón, Victoria; Herrera, Francisco; 33
Sentiment Analysis is an extensively studied task, however an important aspect yet to study is the underlying structural information of opinions. An important aspect to tackle is the analysis underlying structural information of opinions. Social media is a great source of user opinions, which are structured in most of the cases in two sections: the title and the content or body of the opinion. We claim that the structure of social media opinions has useful information for the polarity classification task. We propose a model for optimizing the contribution of that underlying structural information for polarity classification. Our model is built by weighting the contribution of each section, title and body. We develop a modified Support Vector Machine that includes a weight parameter, which is optimized via a line-search strategy. We evaluate our proposal on three datasets of reviews from different domains written in two different versions of the Spanish language. The results show that our model outperforms the classification of the joint or individual classification of each section of the opinion. Therefore, our claim holds.
33
(33, 33) Maldonado, Sebastián; Saltos, Ramiro; Vairetti, Carla; Delpiano, José; 33
Dataset shift is a relevant topic in unsupervised learning since many applications face evolving environments, causing an important loss of generalization and performance. Most techniques that deal with this issue are designed for data stream clustering, whose goal is to process sequences of data efficiently under Big Data. In this study, we claim dataset shift is an issue for static clustering tasks in which data is collected over a long period. To mitigate it, we propose Time-weighted kernel k-means, a k-means variant that includes a time-dependent weighting process. We do this via the induced ordered weighted average (IOWA) operator. The weighting process acts as a gradual forgetting mechanism, prioritizing recent examples over outdated ones in the clustering algorithm. The computational experiments show the potential Time-weighted kernel k-means has in evolving environments.
33
(33, 33) Irarrázaval, María Elisa; Maldonado, Sebastián; Pérez, Juan Eduardo; Vairetti, Carla Marina; 33
Traffic pumping is a type of fraud committed in several countries, in which small telephone operators inflate the number of incoming calls to their networks, profiting from a higher access charge in relation to the network operator associated with the origin of the call. The identification of traffic pumping is complex due to the lack of labels for performing supervised learning, and the scarce literature on the topic. We propose a decision support system for fraud detection via clustering and decision trees. After data collection and feature engineering, we group the potential fraud cases into various clusters via an unsupervised learning approach. Then, we constructed a decision tree by using the cluster memberships as labels, evolving into the rules of a given variable and a certain label required for filing lawsuits against the suspicious cases. Telecommunication experts validate these rules to seek a legal resource against alleged perpetrators. We present the results of a case study from a Chilean telecommunication provider. All the lawsuits taken by the legal department were granted, confirming our success in dramatically reducing current and future fraud losses for the company.
33
(33, 33) De Bock, Koen W.; Coussement, Kristof; Caigny, Arno De; S?owi?ski, Roman; Baesens, Bart; Boute, Robert N.; Choi, Tsan Ming; Delen, Dursun; Kraus, Mathias; Lessmann, Stefan; Maldonado, Sebastián; Martens, David; Óskarsdóttir, María; Vairetti, Carla; Verbeke, Wouter; Weber, Richard; 33
The ability to understand and explain the outcomes of data analysis methods, with regard to aiding decision-making, has become a critical requirement for many applications. For example, in operational research domains, data analytics have long been promoted as a way to enhance decision-making. This study proposes a comprehensive, normative framework to define explainable artificial intelligence (XAI) for operational research (XAIOR) as a reconciliation of three subdimensions that constitute its requirements: performance, attributable, and responsible analytics. In turn, this article offers in-depth overviews of how XAIOR can be deployed through various methods with respect to distinct domains and applications. Finally, an agenda for future XAIOR research is defined.
Analytics-driven complaint prioritisation via deep learning and multicriteria decision-making
Vairetti, Carla; Aránguiz, Ignacio; Maldonado, Sebastián; Karmy, Juan Pablo; Leal, Alonso
The Cobb-Douglas Learning Machine
Maldonado, Sebastián; López, Julio; Carrasco, Miguel