Ricardo B. C. Prudêncio

Lucas F. F. Cardoso, José de S. Ribeiro, Vitor C. A. Santos et al.

Intelligent systems that use Machine Learning classification algorithms are increasingly common in everyday society. However, many systems use black-box models that do not have characteristics that allow for self-explanation of their predictions. This situation leads researchers in the field and society to the following question: How can I trust the prediction of a model I cannot understand? In this sense, XAI emerges as a field of AI that aims to create techniques capable of explaining the decisions of the classifier to the end-user. As a result, several techniques have emerged, such as Explanation-by-Example, which has a few initiatives consolidated by the community currently working with XAI. This research explores the Item Response Theory (IRT) as a tool to explaining the models and measuring the level of reliability of the Explanation-by-Example approach. To this end, four datasets with different levels of complexity were used, and the Random Forest model was used as a hypothesis test. From the test set, 83.8% of the errors are from instances in which the IRT points out the model as unreliable.

Manuel Ferreira-Junior, Jessica T. S. Reinaldo, Telmo M. Silva Filho et al.

Item response theory aims to estimate respondent's latent skills from their responses in tests composed of items with different levels of difficulty. Several models of item response theory have been proposed for different types of tasks, such as binary or probabilistic responses, response time, multiple responses, among others. In this paper, we propose a new version of $β^3$-IRT, called $β^{4}$-IRT, which uses the gradient descent method to estimate the model parameters. In $β^3$-IRT, abilities and difficulties are bounded, thus we employ link functions in order to turn $β^{4}$-IRT into an unconstrained gradient descent process. The original $β^3$-IRT had a symmetry problem, meaning that, if an item was initialised with a discrimination value with the wrong sign, e.g. negative when the actual discrimination should be positive, the fitting process could be unable to recover the correct discrimination and difficulty values for the item. In order to tackle this limitation, we modelled the discrimination parameter as the product of two new parameters, one corresponding to the sign and the second associated to the magnitude. We also proposed sensible priors for all parameters. We performed experiments to compare $β^{4}$-IRT and $β^3$-IRT regarding parameter recovery and our new version outperformed the original $β^3$-IRT. Finally, we made $β^{4}$-IRT publicly available as a Python package, along with the implementation of $β^3$-IRT used in our experiments.

Lucas Cardoso, Vitor Santos, José Ribeiro et al.

Benchmarking is a fundamental practice in machine learning (ML) for comparing the performance of classification algorithms. However, traditional evaluation methods often overlook a critical aspect: the joint consideration of dataset complexity and an algorithm's ability to generalize. Without this dual perspective, assessments may favor models that perform well on easy instances while failing to capture their true robustness. To address this limitation, this study introduces a novel evaluation methodology that combines Item Response Theory (IRT) with the Glicko-2 rating system, originally developed to measure player strength in competitive games. IRT assesses classifier ability based on performance over difficult instances, while Glicko-2 updates performance metrics - such as rating, deviation, and volatility - via simulated tournaments between classifiers. This combined approach provides a fairer and more nuanced measure of algorithm capability. A case study using the OpenML-CC18 benchmark showed that only 15% of the datasets are truly challenging and that a reduced subset with 50% of the original datasets offers comparable evaluation power. Among the algorithms tested, Random Forest achieved the highest ability score. The results highlight the importance of improving benchmark design by focusing on dataset quality and adopting evaluation strategies that reflect both difficulty and classifier proficiency.

Kecia G. Moura, Ricardo B. C. Prudêncio, George D. C. Cavalcanti

Label noise detection has been widely studied in Machine Learning because of its importance in improving training data quality. Satisfactory noise detection has been achieved by adopting ensembles of classifiers. In this approach, an instance is assigned as mislabeled if a high proportion of members in the pool misclassifies it. Previous authors have empirically evaluated this approach; nevertheless, they mostly assumed that label noise is generated completely at random in a dataset. This is a strong assumption since other types of label noise are feasible in practice and can influence noise detection results. This work investigates the performance of ensemble noise detection under two different noise models: the Noisy at Random (NAR), in which the probability of label noise depends on the instance class, in comparison to the Noisy Completely at Random model, in which the probability of label noise is entirely independent. In this setting, we investigate the effect of class distribution on noise detection performance since it changes the total noise level observed in a dataset under the NAR assumption. Further, an evaluation of the ensemble vote threshold is conducted to contrast with the most common approaches in the literature. In many performed experiments, choosing a noise generation model over another can lead to different results when considering aspects such as class imbalance and noise level ratio among different classes.

Lucas F. F. Cardoso, Vitor C. A. Santos, Regiane S. Kawasaki Francês et al.

The experiments covered by Machine Learning (ML) must consider two important aspects to assess the performance of a model: datasets and algorithms. Robust benchmarks are needed to evaluate the best classifiers. For this, one can adopt gold standard benchmarks available in public repositories. However, it is common not to consider the complexity of the dataset when evaluating. This work proposes a new assessment methodology based on the combination of Item Response Theory (IRT) and Glicko-2, a rating system mechanism generally adopted to assess the strength of players (e.g., chess). For each dataset in a benchmark, the IRT is used to estimate the ability of classifiers, where good classifiers have good predictions for the most difficult test instances. Tournaments are then run for each pair of classifiers so that Glicko-2 updates performance information such as rating value, rating deviation and volatility for each classifier. A case study was conducted hereby which adopted the OpenML-CC18 benchmark as the collection of datasets and pool of various classification algorithms for evaluation. Not all datasets were observed to be really useful for evaluating algorithms, where only 10% were considered really difficult. Furthermore, the existence of a subset containing only 50% of the original amount of OpenML-CC18 was verified, which is equally useful for algorithm evaluation. Regarding the algorithms, the methodology proposed herein identified the Random Forest as the algorithm with the best innate ability.

Tarsicio Lucas, Teresa Ludermir, Ricardo Prudencio et al.

The optimization of Artificial Neural Networks (ANNs) is an important task to the success of using these models in real-world applications. The solutions adopted to this task are expensive in general, involving trial-and-error procedures or expert knowledge which are not always available. In this work, we investigated the use of meta-learning to the optimization of ANNs. Meta-learning is a research field aiming to automatically acquiring knowledge which relates features of the learning problems to the performance of the learning algorithms. The meta-learning techniques were originally proposed and evaluated to the algorithm selection problem and after to the optimization of parameters for Support Vector Machines. However, meta-learning can be adopted as a more general strategy to optimize ANN parameters, which motivates new efforts in this research direction. In the current work, we performed a case study using meta-learning to choose the number of hidden nodes for MLP networks, which is an important parameter to be defined aiming a good networks performance. In our work, we generated a base of meta-examples associated to 93 regression problems. Each meta-example was generated from a regression problem and stored: 16 features describing the problem (e.g., number of attributes and correlation among the problem attributes) and the best number of nodes for this problem, empirically chosen from a range of possible values. This set of meta-examples was given as input to a meta-learner which was able to predict the best number of nodes for new problems based on their features. The experiments performed in this case study revealed satisfactory results.

Lucas F. F. Cardoso, Vitor C. A. Santos, Regiane S. K. Francês et al.

Despite the availability of benchmark machine learning (ML) repositories (e.g., UCI, OpenML), there is no standard evaluation strategy yet capable of pointing out which is the best set of datasets to serve as gold standard to test different ML algorithms. In recent studies, Item Response Theory (IRT) has emerged as a new approach to elucidate what should be a good ML benchmark. This work applied IRT to explore the well-known OpenML-CC18 benchmark to identify how suitable it is on the evaluation of classifiers. Several classifiers ranging from classical to ensembles ones were evaluated using IRT models, which could simultaneously estimate dataset difficulty and classifiers' ability. The Glicko-2 rating system was applied on the top of IRT to summarize the innate ability and aptitude of classifiers. It was observed that not all datasets from OpenML-CC18 are really useful to evaluate classifiers. Most datasets evaluated in this work (84%) contain easy instances in general (e.g., around 10% of difficult instances only). Also, 80% of the instances in half of this benchmark are very discriminating ones, which can be of great use for pairwise algorithm comparison, but not useful to push classifiers abilities. This paper presents this new evaluation methodology based on IRT as well as the tool decodIRT, developed to guide IRT estimation over ML benchmarks.

Yu Chen, Telmo Silva Filho, Ricardo B. C. Prudêncio et al.

Item Response Theory (IRT) aims to assess latent abilities of respondents based on the correctness of their answers in aptitude test items with different difficulty levels. In this paper, we propose the $β^3$-IRT model, which models continuous responses and can generate a much enriched family of Item Characteristic Curve (ICC). In experiments we applied the proposed model to data from an online exam platform, and show our model outperforms a more standard 2PL-ND model on all datasets. Furthermore, we show how to apply $β^3$-IRT to assess the ability of machine learning classifiers. This novel application results in a new metric for evaluating the quality of the classifier's probability estimates, based on the inferred difficulty and discrimination of data instances.