While Automated Machine Learning (AutoML) has seen widespread adoption in supervised learning, its application to clustering remains partially investigated. Existing scenarios for unsupervised learning, especially clustering, often focus on algorithm selection and hyperparameter optimisation only, with limited support for full pipeline synthesis. As clustering quality is highly sensitive to appropriate selection of preprocessing steps, ignoring pipeline optimisation heavily affects performance. To address this gap, we propose an evolutionary optimisation framework for the automated synthesis of complete clustering pipelines, including preprocessing steps, by extending the Tree-based Optimisation Tool (TPOT) originally designed for supervised tasks. The framework optimises pipelines using a combination of Clustering Validity Indices (CVIs) and meta-learning. We evaluated our framework on 25 benchmark clustering datasets and our results highlight two key factors for effective clustering automation: the integration of meta-features into the optimisation process and the inclusion of preprocessing steps in the pipeline.

Evolving multimedia systems are increasingly being adopted in virtual reality and gaming applications. Such systems emphasize immersion to engage users by bridging the gap between real and virtual content. In this context, visual and acoustic stimuli are the two key media that dictate such immersion. While visual 3D rendering is advancing rapidly, the same is not true for audio, where most research is limited to the reconstruction of the room impulse response (RIR) using omnidirectional audio or, at best, binaural. Such methods do not adequately account for the directions and orientations of the acoustic signals with respect to either the source or the listener, thereby compromising immersion quality. In this work, we explore the effect of adding such "directionality" to the training data to improve the estimation of the room’s acoustic parameters. A more accurate set of such parameters implies in fact a more realistic predicted RIR, leading to a more immersive experience of the acoustic scene. Specifically, we propose a novel framework driven by a suitable loss function to account for directionality in ambisonic microphones, and novel variants of loss functions for both omnidirectional and ambisonic cases. We also propose to account for microphone characteristics and their contribution to the predicted RIRs. Experiments were performed using two datasets of real recordings and the results established the efficacy of the proposed methods

Signal processing can be applied in the biomedical field, using electromyography (EMG) signals to analyze muscle fatigue, which can be defined as the evidence of loss of muscle strength during muscle contraction. Besides many works of muscle fatigue, there is still a need to know which are the best techniques, especially in dynamic contractions. This contraction is more complex than isometric because it includes body movement, contraction, and muscle rest. The main contribution of this paper is the analysis of the spectral features from EMG signals recorded during dynamic contractions. The features from the frequency domain were extracted from the EMG signal and were made the classification between “fatigue” and “no fatigue”. The database contains signals from three muscles of the quadriceps femoris muscles of the right leg of 17 volunteers, totaling 51 signals with 30 seconds captured during a pedaling activity. After the classification, was observed the use or not of all proposed features, and were made a local analysis of the impact of each feature in the classification. From the initially considered 17 features, 12 were used for classification, obtaining an 89% accuracy rate.

Encoding methods are employed across several process mining tasks, including predictive process monitoring, anomalous case detection, trace clustering, etc. These methods are usually performed as preprocessing steps and are responsible for mapping complex event data information into a numerical feature space. Most papers choose existing encoding methods arbitrarily or employ a strategy based on expert domain knowledge. Moreover, existing methods are employed by using their default parameters without evaluating other options. This practice can lead to several drawbacks, such as suboptimal performance and unfair comparisons with the state-of-the-art. Therefore, this work aims at providing a comprehensive survey and benchmark on event log encoding by comparing 27 methods, from different natures, in terms of expressivity, scalability, correlation, and domain agnosticism. To the best of our knowledge, this is the most comprehensive study so far focusing on trace encoding in process mining. It contributes to maturing awareness about the role of trace encoding in process mining pipelines and sheds light on issues, concerns, and future research directions regarding the use of encoding methods to bridge the gap between machine learning models and process mining.

Supervised data stream learning depends on the incoming sample’s true label to update a classifier’s model. In real life, obtaining the ground truth for each instance is a challenging process; it is highly costly and time consuming. Active Learning has already bridged this gap by finding a reduced set of instances to support the creation of a reliable stream classifier. However, identifying a reduced number of informative instances to support a suitable classifier update and drift adaptation is very tricky. To better adapt to concept drifts using a reduced number of samples, we propose an online tuning of the Uncertainty Sampling threshold using a meta-learning approach. Our approach exploits statistical meta-features from adaptive windows to meta-recommend a suitable threshold to address the trade-off between the number of labelling queries and high accuracy. Experiments exposed that the proposed approach provides the best trade-off between accuracy and query reduction by dynamic tuning the uncertainty threshold using lightweight meta-features.

Online process mining refers to a class of techniques for analyzing in real-time event streams generated by the execution of business processes. These techniques are crucial in the reactive monitoring of business processes, timely resource allocation and detection/prevention of dysfunctional behavior. Many interesting advances have been made by the research community in recent years, but there is no consensus on the exact set of properties these techniques have to achieve. This paper tills the gap by identitying a set of evaluation goals for online process mining and examining their fulfillment in the state of the art. We discuss parameters and techniques regulating the balance between conflicting goals and outline research needed for their improvement. Concept drift detection is crucial in this sense but, as demonstrated by our experiments, it is only partially supported by current solutions.

Choosing the most suitable algorithm to perform a machine learning task for a new problem is a recurrent and complex task. In multi-target regression tasks, when problem transformation methods are applied, this choice is even harder. The reason is the need to simultaneously choose the problem transformation method and the base learning algorithm. This work investigates how to bridge the gap of method/base learner recommendation for problems with multiple outputs. In meta-learning experiments, we use a large number of multi-target regression datasets to investigate whether using meta-learning can provide good recommendations. To do this, we compared the meta-models induced by 3 different ML algorithms, including three variations for each of them, and selected 58 meta-features that we believe are relevant for extracting good dataset descriptions for the meta-learning process. In the experimental results, the meta-models outperformed the baselines (Majority and Random) by recommending the most suitable solution for multi-target regression (for the transformation method and base-learner) with high predictive performance, including real-world applications. The meta-features and the relation between the transformation method and base-learner provided important insights regarding the optimal problem transformation method. Furthermore, when comparing the application of algorithm adaptation and problem transformation methods, our meta-learning proposal was capable of statistically overcoming all competitors, which resulted in a predictive performance using the best choice per problem.

Abstract Cocoa hybridisation generates new varieties which are resistant to several plant diseases, but has individual chemical characteristics that affect chocolate production. Image analysis is a useful method for visual discrimination of cocoa beans, while deep learning (DL) has emerged as thede factotechnique for image processing . However, these algorithms require a large amount of data and careful tuning of hyperparameters. Since it is necessary to acquire a large number of images to encompass the wide range of agricultural products, in this paper, we compare a Deep Computer Vision System (DCVS) and a traditional Computer Vision System (CVS) to classify cocoa beans into different varieties. For DCVS, we used a Resnet18 and Resnet50 as backbone, while for CVS, we experimented traditional machine learning algorithms, Support Vector Machine (SVM), and Random Forest (RF). All the algorithms were selected since they provide good classification performance and their potential application for food classification A dataset with 1,239 samples was used to evaluate both systems. The best accuracy was 96.82% for DCVS (ResNet 18), compared to 85.71% obtained by the CVS using SVM. The essential handcrafted features were reported and discussed regarding their influence on cocoa bean classification. Class Activation Maps was applied to DCVS’s predictions, providing a meaningful visualisation of the most important regions of the images in the model.

Online process mining refers to a class of techniques for analyzing in real-time event streams generated by the execution of business processes. These techniques are crucial in the reactive monitoring of business processes, timely resource allocation and detection/prevention of dysfunctional behavior. Many interesting advances have been made by the research community in recent years, but there is no consensus on the exact set of properties these techniques have to achieve. This article fills the gap by identifying a set of evaluation goals for online process mining and examining their fulfillment in the state of the art. We discuss parameters and techniques regulating the balance between conflicting goals and outline research needed for their improvement. Concept drift detection is crucial in this sense but, as demonstrated by our experiments, it is only partially supported by current solutions.

Assuring anomaly-free business process executions is a key challenge for many organizations. Traditional techniques address this challenge using prior knowledge about anomalous cases that is seldom available in real-life. In this work, we propose the usage of word2vec encoding and One-Class Classification algorithms to detect anomalies by relying on normal behavior only. We investigated 6 different types of anomalies over 38 real and synthetics event logs, comparing the predictive performance of Support Vector Machine, One-Class Support Vector Machine, and Local Outlier Factor. Results show that our technique is viable for real-life scenarios, overcoming traditional machine learning for a wide variety of settings where only the normal behavior can be labeled.

Edge computing (EC) is a promising technology capable of bridging the gap between Cloud computing services and the demands of emerging technologies such as the Internet of Things (IoT). Most EC-based solutions, from wearable devices to smart cities architectures, benefit from Machine Learning (ML) methods to perform various tasks, such as classification. In these cases, ML solutions need to deal efficiently with a huge amount of data, while balancing predictive performance, memory and time costs, and energy consumption. The fact that these data usually come in the form of a continuous and evolving data stream makes the scenario even more challenging. Many algorithms have been proposed to cope with data stream classification, e.g., Very Fast Decision Tree (VFDT) and Strict VFDT (SVFDT). Recently, Online Local Boosting (OLBoost) has also been introduced to improve predictive performance without modifying the underlying structure of the decision tree produced by these algorithms. In this work, we compared the four-way relationship among time efficiency, energy consumption, predictive performance, and memory costs, tuning the hyperparameters of VFDT and the two versions of SVFDT with and without OLBoost. Experiments over 6 benchmark datasets using an EC device revealed that VFDT and SVFDT-I were the most energy-friendly algorithms, with SVFDT-I also significantly reducing memory consumption. OLBoost, as expected, improved the predictive performance, but caused a deterioration in memory and energy consumption.

Image segmentation is a key issue in image processing. New image segmentation algorithms have been proposed in the last years. However, there is no optimal algorithm for every image processing task. The selection of the most suitable algorithm usually occurs by testing every possible algorithm or using knowledge from previous problems. These processes can have a high computational cost. Meta-learning has been successfully used in the machine learning research community for the recommendation of the most suitable machine learning algorithm for a new dataset. We believe that meta-learning can also be useful to select the most suitable image segmentation algorithm. This hypothesis is investigated in this paper. For such, we perform experiments with eight segmentation algorithms from two approaches using a segmentation benchmark of 300 images and 2100 augmented images. The experimental results showed that meta-learning can recommend the most suitable segmentation algorithm with more than 80% of accuracy for one group of algorithms and with 69% for the other group, overcoming the baselines used regarding recommendation and segmentation performance.

Transposable Elements (TEs) are DNA sequences capable of moving within a cell's genome. Their transposition has many effects in genomes, such as creating genetic variability and promoting changes in genes' functionality. Recently, TEs classification has been addressed using Machine Learning (ML), more specifically by Hierarchical Classification (HC) methods. Such works proved to be superior than previous ones in the literature. However, there is still room for improvement performance wise. In this direction, Deep Neural Networks (DNNs) have attracted a lot of attention in ML. In particular, Stacked Denoising Auto-Encoders (DAEs) and Deep Multi Layer-Perceptrons (MLPs) are known to provide outstanding results. By performing an extensive evaluation, our results point out that DNNs can enhance the performance of HC methods, being able to push further the state-of-art in TEs' classification.

With the ageing and growth of the population, some chronic diseases, such as Parkinson's disease (PD), urge the society to a health-conscious looking for better health system designs. Some recent research endeavour has been supported by solutions grounded in ubiquitous healthcare (u-Health) coupling telemedicine, context awareness and decision support capabilities. In this work, we propose a u-healthcare system to pre-diagnose PD based on the speech signal of people under voice call. The speech stream is sampled as well as processed to support the pre-diagnose using machine learning (ML). Experiments were conducted over a PD voice dataset composed of 40 individuals by using five different ML algorithms. Based on a linear Support Vector Machine (SVM) model, a false negative rate of 10% was obtained when classifying the locution of number "three".

Social interactions take place in environments that influence people’s behaviours and perceptions. Nowadays, the users of Online Social Network (OSN) generate a massive amount of content based on social interactions. However, OSNs wide popularity and ease of access created a perfect scenario to practice malicious activities, compromising their reliability. To detect automatic information broadcast in OSN, we developed a wavelet-based model that classifies users as being human, legitimate robot, or malicious robot, as a result of spectral patterns obtained from users’ textual content. We create the feature vector from the Discrete Wavelet Transform along with a weighting scheme called Lexicon-based Coefficient Attenuation. In particular, we induce a classification model using the Random Forest algorithm over two real Twitter datasets. The corresponding results show the developed model achieved an average accuracy of 94.47% considering two different scenarios: single theme and miscellaneous one.

Botnets have been a serious threat to the Internet security. With the constant sophistication and the resilience of them, a new trend has emerged, shifting botnets from the traditional desktop to the mobile environment. As in the desktop domain, detecting mobile botnets is essential to minimize the threat that they impose. Along the diverse set of strategies applied to detect these botnets, the ones that show the best and most generalized results involve discovering patterns in their anomalous behavior. In the mobile botnet field, one way to detect these patterns is by analyzing the operation parameters of this kind of applications. In this paper, we present an anomaly-based and host-based approach to detect mobile botnets. The proposed approach uses machine learning algorithms to identify anomalous behaviors in statistical features extracted from system calls. Using a self-generated dataset containing 13 families of mobile botnets and legitimate applications, we were able to test the performance of our approach in a close-to-reality scenario. The proposed approach achieved great results, including low false positive rates and high true detection rates.

Hierarchical Classification (HC) consists of classification problems whose classes are structured in a hierarchical fashion. Many problems are addressed by HC, in special a decent amount of works dealt with bioinformatics related problems such as Protein Function Prediction (PFP) and Transposable Elements (TEs) classification. Both of them are still a challenging task for HC due to the noisy and imbalanced nature of the datasets. As a countermeasure, Stacking is an ensemble method capable of generalizing knowledge from many classifiers. In this work, we propose three Stacking methods for HC and evaluate its performance on PFP and TEs datasets. Our results show that, when compared to regular Stacking and state-of-art methods from the literature, our methods are able to obtain superior or competitive performances.

We developed a wavelet-based approach for account classification that detects textual dissemination by bots on an Online Social Network (OSN). Its main objective is to match account patterns with humans, cyborgs or robots, improving the existing algorithms that automatically detect frauds. With a computational cost suitable for OSNs, the proposed approach analyses the distribution of key terms. The descriptors, a wavelet-based feature vector for each user's account, work in conjunction with a new weighting scheme, called Lexicon Based Coefficient Attenuation (LBCA) and serve as inputs to one of the classifiers tested: Random Forests and Multilayer Perceptrons. Experiments were performed using a set of posts crawled during the 2014 FIFA World Cup, obtaining accuracies within the range from 94 to 100%. (C) 2015 Elsevier Inc. All rights reserved.

Misleading diagnosis of skin diseases may result in complications during the healing process. Skin images provide an important contribution to medical staff on storing and exchanging information to try preventing misdiagnosis. For such, image segmentation process may benefit from use of machine learning techniques, increasing simplicity of procedure, reducing computational costs and improving the diagnosis. This paper presents a comparison among different paradigms of machine learning to validate the segmentation of medical images of lower members ulcers, this segmentation allows wound pattern recognition to determinate injury region aiming at reducing the subjectivity of human evaluation.

This paper presents a new tool for anomaly detection called Correlational Paraconsistent Machine (CPM), which is applied in mathematical treatment of uncertainties that may arise during the normal network traffic behavior modeling. The presented CPM incorporates two unsupervised models for traffic characterization, and principles on paraconsistency to evaluate the network for the presence of irregularity at traffic levels. Using flow data collected at the backbone of a real network, we present two case studies and show that our approach can accurately detect anomalies and validate the consistency of the process.

This paper presents a seed finding method for region growing segmentation approach using color channel energy in image regions. Instead of using the RGB system separated for each pixel, the proposal uses the energy on each color channel to improve the range of the possible values, then creates a more specific seed to detail different regions. Region size used to calculate energy was adjusted to verify the proposed method. Images used were real wound photos, taken from patients undergoing treatment at the university hospital. Results showed that energy on regions presents enough information to segment, leading to a high percentage of matching with experts marks.

Detecting anomalies accurately in network traffic behavior is essential for a variety of network management and security tasks. This paper presents an anomaly detection approach employing Digital Signature of Network Segment using Flow Analysis (DSNSF), generated with an ARIMA model. Also, a functional algorithm based on a non-classical logic called Paraconsistent Logic is proposed aiming to avoid high false alarms rates. The key idea of the proposed approach is to characterize the normal behavior of network traffic and then identify the traffic patterns behavior that might harm networks services. Experimental results on a real network demonstrate the effectiveness the proposed approach. The results are promising, showing that the flow analysis performed is able to detect anomalous traffic with precision, sensitivity and good performance.

Discriminative training of Gaussian Mixture Models (GMMs) for speech or speaker recognition purposes is usually based on the gradient descent method, in which the iteration step-size, epsilon, uses to be defined experimentally. In this letter, we derive an equation to adaptively determine epsilon, by showing that the second-order Newton-Raphson iterative method to find roots of equations is equivalent to the gradient descent algorithm.

In this paper, two independent Support Vector Machines were connected to a paraconsistent logic unit in order to establish a new classification scheme which takes into account the degrees of faith and uncertainty of a certain statement. By using this approach, one can classify an input signal as matching one of two independent classes or both of them. In our experiments, speech data constitute the classification elements which were adopted, and the results demonstrate the efficacy of the proposed approach.

This paper introduces the paradigm of joint time- frequency-shape analysis of discrete-time signals by means of a novel transform called discrete shapelet transform (DST). This approach allows one to And the time-support of frequencies, and, at the same time, analyze a shape. Applications are possible for a diversity of fields in signal processing, and the results assure the efficacy of the proposed transform.

Nowadays, fractal analysis has been successfully applied to digital speech processing, particularly for word and phoneme segmentation, which represents one of the fundamental steps in automatic speech recognition systems. The practical use of fractal analysis for this purpose should match two principles: low computational cost, to allow the use in real-time, and accuracy in the results, in order to produce a satisfactory segmentation, sending the correct data to the classifier. Aiming at meeting these two requirements, this work proposes a technique for speech segmentation based on the fractal dimension, which is obtained by using the discrete wavelet transform that avoids the use of 1/k pre-filtering. Many families of wavelets are presented and compared, and the results assure the efficacy of the proposed method.

Towards an optimization-oriented approach for audio coding, this paper presents improved rate-distortion and perceptual strategies for bit allocation. The algorithm is based on best basis wavelet-packet trees and fractal dimension calculation. Transparent coding of high quality audio, signals sampled at 44.1 KHz using 16 bits PCM, is effectively achieved at low bit rates. Real time working of the decoder is confirmed, reassuring the usability of the proposed technique