Automated recommendation of machine learning algorithms is receiving a large deal of attention, not only because they can recommend the most suitable algorithms for a new task, but also because they can support efficient hyper-parameter tuning, leading to better machine learning solutions. The automated recommendation can be implemented using meta-learning, learning from previous learning experiences, to create a meta-model able to associate a data set to the predictive performance of machine learning algorithms. Although a large number of publications report the use of meta-learning, reproduction and comparison of meta-learning experiments is a difficult task. The literature lacks extensive and comprehensive public tools that enable the reproducible investigation of the different meta-learning approaches. An alternative to deal with this difficulty is to develop a meta-feature extractor package with the main characterization measures, following uniform guidelines that facilitate the use and inclusion of new meta-features. In this paper, we propose two Meta-Feature Extractor (MFE) packages, written in both Python and R, to fill this lack. The packages follow recent frameworks for meta-feature extraction, aiming to facilitate the reproducibility of meta-learning experiments.

Epilepsy is the fourth most incident neurological disorder, for which approximately 50 million people are affected. The epilepsy is characterized by repeated seizures, occasioned by temporary electrical disturbances of the brain, and it is usually evidenced in four distinct events: pre-ictal, ictal, post-ictal, and interictal. This disorder can be diagnosed by electroencephalogram (EEG), which is the result of brain electrical activity monitoring. In the current work, we built multiclass classifiers in order to differentiate EEG segments among normal, interictal, and ictal classes. For this, the multitaper method was computed to generate power spectrum, spectrogram, and bispectrogram from each EEG segment, in which a total of 102 features were extracted. For classifier building, we used all-against-all approach to decompose the multiclass problem into binary subproblems, in which the following machine learning algorithms were applied: random forest, 1-nearest-neighbor, linear discriminant analysis (LDA), backpropagation based on multilayer perceptron (BP-MLP), k-means based on radial basis function network (KM-RBFN), and sequential minimal optimization. In the evaluation by ten-fold stratified cross-validation, the BP-MLP reached the highest accuracy (98.33%). Afterward, an extremely significant statistical difference was found among classifiers. In a post hoc test application, it was proved that the KM-RBFN classifier reached an inferior performance compared to LDA and BP-MLP predictive models. All classifiers built in this work achieved promising results, in which accuracy above 90% was obtained for classification of EEG segments.

An automatic generation of medical report method (AGMedRep) was proposed in order to process electroencephalogram (EEG) segments using machine learning (ML) to generate textual reports for epilepsy detection. This method is applied in two phases: (1) predictive model building, and (2) automatic generation of medical reports. In Phase 1, classifiers are built and evaluated, in which the best predictive model is chosen to perform Phase 2 for report generation. The AGMedRep was applied in a set of 500 EEG segments, 100 for each class, in a total of five classes. In the first phase, 90 signal segments for each class were selected for feature extraction and classifier building. Posteriorly, five different report expression were defined, one for each class. In each EEG segment, all possible combination of cross-correlogram, power spectrum, spectrogram (SG), and bispectrogram (BS) were computed for feature extraction, generating a total of 15 different datasets. Afterwards, ML methods, such as decision tree, 1-nearest-neighbor (1NN), naive Bayes, backpropagation based on multilayer perceptron (MLP), and support vector machine (SVM) were applied in each dataset considering a multiclass decomposition approach, generating a total of 75 (15 datasets × 5 ML techniques) classifiers. In the evaluation step, it was found that the "SG" 1NN, "SG and BS" MLP, and "SG and BS" SVM reached, statistically, the best performances for signal classification, but a statistical difference was not found among them. In this sense, the "SG" 1NN model was chosen for report generation due to its lower computation cost, and it proved to be one of the most accurate models. In the second phase, the 50 remaining EEG segments were equally distributed, randomly, into ten folders in order to simulate individual EEG exams. Finally, the chosen classifier was computed in each examination to construct textual reports. As a result, the 1NN model reached an average of 84% accuracy in the report generation. Our results suggest that the AGMedRep can be able to assist medical experts in the identification of patterns related to epileptic events in EEG.

According to World Health organization, epilepsy is one of the most common chronic neurological disorder, affecting approximately 50 million people. This infirmity presents four kinds of events: pre-ictal, ictal, post-ictal, and interictal. Epilepsy can be diagnosed through electroencephalogram (EEG). The interictal activity, for example, on an EEG, is widely accepted as an epilepsy symptom. However, the differentiation between normal and interictal EEG segments is difficult because they can have similar patterns. Also, EEG from patients with epilepsy can contain normal events. In this work, we built classifiers to differentiate between normal and interictal EEG. Our proposed process was applied in a set of 200 EEG segments. For this, power spectrum (PS) was computed for each signal, and 18 measures were extracted from PS considering five frequency bands: delta, theta, alpha, beta, and entire frequency range (between lower delta and higher beta frequency range). Thus, from each PS, 90 features were extracted. The following machine learning methods were applied to build the classifiers: random forest, INN, naive Bayes (NB), MLP, and SVM. In the evaluation by cross-validation approach, a statistically significant difference was not found among classifiers, whose error values resulted in the p-value of 0.1089 by Friedman test. On the other hand, by confusion matrices and their parameters, it was found that the NB classifier reached the best performance to detect normal EEG segments. For detection of interictal activity, the MLP and SVM classifiers achieved the best results. All classifiers built in this work reached promising results for differentiation between normal and interictal EEG patterns.

Epilepsy is the fourth most frequent neurological disorder. Epileptic seizures are the result of temporary electrical disturbances in the brain. This disorder can be diagnosed by electroencephalograms (EEG). Accordingly, data mining supported by machine learning (ML) methods can be used to find patterns in EEG and to build classifiers. However, the presence of physiological abnormalities is considered rare in medical data. Also, rhythmic activities in an EEG of a healthy person varies in different situations, such as with closed or opened eyes. Other issue is that normal and interictal epileptic EEG can contain similar patterns. In this work, we developed classifiers in order to differentiate EEG of healthy volunteers, with opened and closed eyes, from epileptic EEG. This approach was applied in an EEG set with 500 segments, in which cross-correlation and power spectrum methods were performed for feature extraction. Subsequently, predictive models based on one-class classification were built using the one-class support vector machine (OSVM) technique in addition to traditional ML methods, such as nearest-neighbor, artificial neural network, and two-class support vector machine (TSVM). The polynomial and radial basis function (RBF) kernels were applied in OSVM and TSVM. To enable the use of traditional ML methods, the OCC was converted to two-class classification problem through assignment of negative (abnormal) class to each EEG segment that does not belong to normal (positive) class. In the evaluation, it was found that the OSVM classifier with RBF kernel reached the best values for all confusion matrix parameters, such as accuracy, positive predictive value, sensitivity, and false positive rate.