Abstract
Data imbalance is characterized by a discrepancy in the number of examples per class of a dataset. This phenomenon is known to deteriorate the performance of classifiers, since they are less able to learn the characteristics of the less represented classes. For most imbalanced datasets, the application of sampling techniques improves the classifier's performance. For small datasets, oversampling has been shown to be the most appropriate strategy since it augments the original set of samples. Although several oversampling strategies have been proposed and tested over the years, the work has mostly focused on binary or multi-class tasks. Motivated by medical applications, where there is often an order associated with the classes (increasing likelihood of malignancy, for instance), the present work tests some existing oversampling techniques in ordinal contexts. Moreover, four new oversampling techniques are proposed. Experiments were made both on private and public datasets. Private datasets concern the assessment of response to treatment on oncologic diseases. The 15 public datasets were chosen since they are widely used in the literature. Results show that data balance techniques improve classification results on ordinal imbalanced datasets, even when these techniques are not specifically designed for ordinal problems. With our pipeline, better or equal to published results were obtained for 10 out of the 15 public datasets with improvements upon a decrease of 0.43 on MMAE.

Abstract
Brain-Computer Interface systems based on motor imagery are able to identify an individual's intent to initiate control through the classification of encephalography patterns. Correctly classifying such patterns is instrumental and strongly depends in a robust machine learning block that is able to properly process the features extracted from a subject's encephalograms. The main objective of this work is to provide an overall view on machine learning stages, aiming to answer the following question: "What are the steps in the classification process that we should worry about?". The obtained results suggest that future research in the field should focus on two main aspects: exploring techniques for dimensionality reduction, in particular, supervised linear approaches, and evaluating adequate validation schemes to allow a more precise interpretation of results.

Abstract
Missing data consists in the lack of information in a dataset and since it directly influences classification performance, neglecting it is not a valid option. Over the years, several studies presented alternative imputation strategies to deal with the three missing data mechanisms, Missing Completely At Random, Missing At Random and Missing Not At Random. However, there are no studies regarding the influence of all these three mechanisms on the latest high-performance Artificial Intelligence techniques, such as Deep Learning. The goal of this work is to perform a comparison study between state-of-the-art imputation techniques and a Stacked Denoising Autoencoders approach. To that end, the missing data mechanisms were synthetically generated in 6 different ways; 8 different imputation techniques were implemented; and finally, 33 complete datasets from different open source repositories were selected. The obtained results showed that Support Vector Machines imputation ensures the best classification performance while Multiple Imputation by Chained Equations performs better in terms of imputation quality. © Springer Nature Switzerland AG 2018.

Abstract
A comparison between co-training and self-training method for single-target regression based on multiples learners is performed. Data streaming systems can create a significant amount of unlabeled data which is caused by label assignment impossibility, high cost of labeling or labeling long duration tasks. In supervised learning, this data is wasted. In order to take advantaged from unlabeled data, semi-supervised approaches such as Co-training and Self-training have been created to benefit from input information that is contained in unlabeled data. However, these approaches have been applied to classification and batch training scenarios. Due to these facts, this paper presents a comparison between Co-training and Self-learning methods for single-target regression in data streams. Rules learning is used in this context since this methodology enables to explore the input information. The experimental evaluation consisted of a comparison between the real standard scenario where all unlabeled data is rejected and scenarios where unlabeled data is used to improve the regression model. Results show evidences of better performance in terms of error reduction and in high level of unlabeled examples in the stream. Despite this fact, the improvements are not expressive.

Abstract
In a single-target regression context, some important systems based on data streaming produce huge quantities of unlabeled data (without output value), of which label assignment may be impossible, time consuming or expensive. Semi-supervised methods, that include the co-training approach, were proposed to use the input information of the unlabeled examples in the improvement of models and predictions. In the literature, the co-training methods are essentially applied to classification and operate in batch mode. Due to these facts, this work proposes a co-training online algorithm for single-target regression to perform model improvement with unlabeled data. This work is also the first-step for the development of online multi-target regressor that create models for multiple outputs simultaneously. The experimental framework compared the performance of this method, when it rejects unalabeled data and when it uses unlabeled data with different parametrization in the training. The results suggest that the co-training method regressor predicts better when a portion of unlabeled examples is used. However, the prediction improvements are relatively small. © Springer International Publishing AG 2017.

Abstract
Given the large volumes and dynamics of data that recommender systems currently have to deal with, we look at online stream based approaches that are able to cope with high throughput observations. In this paper we describe work on incremental neighborhood based and incremental matrix factorization approaches for binary ratings, starting with a general introduction, looking at various approaches and describing existing enhancements. We refer to recent work on forgetting techniques and multidimensional recommendation. We will also focus on adequate procedures for the evaluation of online recommender algorithms.