The aluminium processing industry relies on cold rolling as a crucial method for producing materials essential to everyday life. As a consequence of thickness reduction: Strip Breakage, which is one of the main degraders in the aluminum rolling process, arises. In more recent related studies, data were mainly chosen by experts, which can sometimes hide crucial causal factors. Therefore, this work develops Machine Learning models to predict strip breakage, by investigating the relationship between features and the breakage event, using attribute selection techniques to reduce the dimensionality of the problem, comparing classification methods based on trees, such as Decision Trees (DT), Random Forest (RF) and Extra Trees (ET). This work uses the following methodology: first, the models were evaluated for the complete set of data with default initial parameters, then a new evaluation of the models was carried out with the optimized parameters still with high data dimensions, and finally, an evaluation was carried out of the models with the best-selected attributes. The model that performed best was the DT with 21 attributes, with a recall or TPR (True Positive Rate) of 0.863 and an AUC (Compute Area Under of ROC Curve) of 0.926. The models also indicated that there is no unique cause that characterizes the breaks in the aluminium cold mill, but several others, such as the effect of the rolling mill, coil history domain, oil mill and lamination cylinders.

Automated machine learning (AutoML) is an increasingly popular approach to building machine learning (ML) models without the need for extensive human intervention.One key component of AutoML is automated data ingestion, which involves automatically collecting, cleaning, and preparing data for use in ML models.This paper aims to analyze the literature in order to identify how automated data ingestion is being developed in the literature.To achieve this goal, a survey was conducted on the state-of-the-art of automated data ingestion using a method based on a systematic literature review, in order to identify the existing practices.A total of 12 articles were initially found, however, after applying filters, only six of them were ultimately utilized in the research, showing that visual data navigation and validation as well as metadata inference are important features for automated data ingestion focused in AutoML.

Spot welding is one of the most frequently used material joining techniques in the Automotive Industry. Splashes are an anomalous condition of material expulsion that occurs randomly during the process and since it might result in welds with inadequate quality, it should be avoided. This study uses data of the Spot Welding process of a manufacturing unit that uses BOSCH technology and aims to apply Autoencoders considering a supervised learning approach to identify the occurrence of splashes. Additionally, its goal is to verify if the Autoencoder would outperform traditional techniques in this context when employed to identify rarer anomalies, as anticipated by the studies in the literature review. For this reason, the results for datasets with different anomaly rates are evaluated.

Based on the need of prioritization of maintenance activities in a BOSCH Spot Welding process in the automotive industry, this work aims to develop anomalous equipment selection methodologies for assisting it.The first one is proposed based on data exploration by checking every possible set of alarms of the machines.A second one is created using multiple data clustering models in order to identify machines that behave differently from the others for certain time periods.Bayesian networks were also applied to assist the identification of cause-and-effect relationships between the warning and error logs.The clustering method proved effective in identifying anomalies, which were later inspected on the shop floor.

In data mining solutions, the data selection phase plays an essential role in the success of decision-making.The tools that operate at this phase need to cater to each domain's technical and management challenges.Using a Domain-Specific Modeling Language (DSML), we found an alternative to abstract data and simplify the selection for Educational Data Mining (EDM) process.This work presents a graphic DSML to represent the problem.We used a case study methodology and implemented a CASE tool for the language evaluation.We acquired evidence that the proposed language simplifies the data selection phase for EDM because it solves the technical and management challenges addressed to this domain.

Predicting that a student is likely to fail in a course is critical for performing early interventions, prevent dropout and increase performance on distance learning. This work investigates the most promising machine learning model to perform this task using a small (35 samples) dataset that concerns two classes of one undergraduate course subject. The results bring evidence that the implemented ensemble can perform a prediction at the end of the first week of the course, with a mean accuracy of 78%, when presented to unseen data. This paper also investigates the influence of past data on the results of the classifiers by building datasets with different time window configurations.

In the last decade, a large volume of data has emerged from the massive use of Virtual Learning Environments (VLE).The information contained in these data has enabled the evolution of Educational Data Mining (EDM), whose objective is to apply Machine Learning (ML) in educational contexts.However, building accurate and robust ML models requires, in most cases, advanced knowledge in data science.To solve such problems, Automated Machine Learning techniques have been studied, to simplify the repetitive processes of Data Mining.To validate the solution, the database of the Núcleo de Educac ¸ão a Distância da Universidade de Pernambuco was used.In comparison with the classic EDM approaches, the applied technique showed a superior result, obtaining an accuracy of 89% in the student performance classification process.This solution is called the Framework de Minerac ¸ão de Dados Educacionais (FMDEV), whose objective is to allow users to validate and make available ML baselines with greater productivity.The results of the experts' opinions prove that the FMDEV can contribute to the construction of better models of ML.

Process mining has emerged as a new scientific research topic on the interface between process modeling and event data gathering.In the search for process models that best fit to reality, the process discovery approach of creating referential processes from observed behavior.However, despite these methods showing relevant results, when faced with noisy and divergent tendencies they end up producing limited results.This work proposes the application of process discovery technique, combined to cluster technique k-means, to generate new process models, considering its conformance checking measures.The proposed solution is applied to an ad hoc workflow.And as a result, the use of the clustering techniques coupled with process discovery showed significant gains in the generation of process models, unlike the standard approach.

Unstructured data represent as much as 90% of all business-relevant information. In Brazil, the practice of printing official journals dates back to the 19th century. Today more than 200 official journals in circulation, which together accumulate around 1.4 billion publications without textual standard. This work proposes the development of a system for unstructured data mining using a Dynamic Ensemble Selection. JudEasy implements, added in addition to classic text pre-processing methods, a set of twelve DES and a static method for creating categorized textual models for Brazilian of official journals. As results the DES-KL model obtained the highest accuracy rate of 96.81% and exceptional precision of 0.99.

Educational data mining can be used to understand data from educational systems to provide subsidies to assist teachers, tutors and decision makers. In this context, the objective of this work was to develop a model to identify patterns of learning in distance education using Data Mining techniques and features extracted from the Meaningful Learning Theory. Seven experiments were carried out to validate the proposed model, which consisted of collecting and analyzing data about students in the seven periods of the Pedagogy course. As a result, it was possible to explain the behavior of groups of students and to validate the proposed model as an essential resource in assisting the decision-making of teachers, tutors, and managers.

According to the legislation of Brazil's Ministry of Education (MEC), the student assessment in distance learning programs (e-learning) is based on face-to-face exams at an educational center and online activities. The legislation also requires that the face-to-face exams must have the heaviest weight in the final performance. Given this, the present article seeks to question whether this requirement is generating students who make minimal use of the resources offered by the e-learning platforms but still achieve passing grades because of the face-to-face exams weight, thus affecting the effectiveness of distance learning. For such purpose, a model has been defined and validated using the Isolation Forest algorithm to identify these anomalies, after which, the behavior of the students regarding use of the online platform was analyzed.

Up to 50% of the students drop out of school in Brazilian universities. Because of the heterogeneity of individuals, it is difficult to determine which are the main causes of this high percentage of students not finishing their degree. In this paper, we employed the Fuzzy C-Means algorithm on a dataset composed of real-world registers of the Biology Undergraduate course from Brazilian universities. We applied the transactional distance theory to select the set of variables which were utilized in the clustering process. The results indicate that the data is better divided into five groups. We observed that the Fuzzy C-Means generated groups based on how engaged the students are, and, in each group, there are two subgroups: students that drop out and do not drop out the course. The type of analysis presented in this work can generate inputs for the institutions to establish new policies to reduce the dropout rate.

Educational data mining is concerned with developing, researching, and applying automated methods to detect patterns in collections of educational data, gaining insights into and explaining phenomena in this scenario. The present study describes the application of the prediction of educational indicators in the Brazilian scenario through ensemble models. Ensemble models usually result in better accuracy and are more stable than individual techniques, since they combine the prediction of their components by providing a result more robust. The first model we developed combining parametric regression techniques with base-level learners. The second model uses the set of methods found in the literature in a Stacking regression application formed by parametric and non-parametric techniques. We compare these models, and the results indicate a smaller prediction error for our Stacking model in most of the scenarios studied.

The paper aims to create an intelligent model of data analysis in the registry of the secondary energy distribution network.With emphasis on the search for possible inconsistencies that can be only cadastral or really physical.For this, it uses some techniques of data mining giving focus for the detection of anomalies.This study used a private database containing information about the assets that make up the secondary energy distribution network, such as: poles, transformers, disconnectors, among others.The research was developed following all steps presented in the CRISP-DM methodology.To detect the anomalies, it was used algorithms Isolation Forest, DBSCAN and BIRCH.As a result, the three algorithms pointed to a set of specialty anomalies, validated by a specialist, however, Isolation forest was more accurate in the inference of the anomalies.From this study, distribution companies will be able to identify risky or financially problematic situations in advance.

With the increasing interest in developing Learning Analytics tools that can be integrated into the well-known Moodle course management systems nowadays, many tools have already been developed.These tools usually requires the user to know data mining techniques, and also requires time to get mining results from the tools.To address this problem, in this article, we present a structure that uses pre-built data mining through Shiny to quickly obtain results with a focus on visualizing data with graphs, and thus allows the integration of other research, called FMDEV.Guided by the proposed framework, a tool was developed for display the data mining results in a few clicks for Moodle users who wish to have them for day-to-day use and allows users with more experience in data mining to integrate new research.Finally, we used FMDEV tool to generate some experimental results using a set of real-life sample data from undergraduate students.

This paper presents an analysis using hierarchical grouping method (ward grouping) and the non-hierarchical grouping method (k-means) to analyze the participation levels in activities and interactions in a virtual forum. Data came from a MOOC and it was on grammatical rules of Brazilian Portuguese. About 5100 participants integrated the course. It lasted about three months and the activities developed by means of the Openredu learning platform. We analyzed data of collaboration, interaction, and discussions in forums, access data and activity on the Openredu. The results pointed to three distinct engagement strategies. Those categories oriented the proposition of an interface design guidelines for MOOC to conceive adaptive strategies that permit to increase engagement and favor an improved learning experience.

This study presents a proposal for the analysis of student's engagement level in an online course. Data was used from a graduate course at a Brazilian public university. The method was based on the process of Educational Data Mining (EDM) to identify transactional distance constructs in the data collected and metrics defined by Social Network Analysis (SNA) and also the use of logistic regression to obtain representative engagement models and factors that affect them. The results stated the most relevant aspects of students' engagement in the course and therefore they indicated on which factors could be made interventions to increase this engagement.

Discussion forums are an important feature in the Distance Education courses, supporting learning and facilitating interaction between students and teachers.This work aims to investigate behavioral aspects of students in virtual learning environments using Educational Data Mining.It is proposed to use the K-Means clustering technique as a way to identify students with common patterns of behavior based on their interactions in the forums.This work achieves good results from the application of clustering technique for this particular issue.

Software Product Line is a software development paradigm created to meet different market segments.This paradigm has shown great acceptance in the corporate environment (Motorola, Nokia, and Hewlett Packard) to allow the construction of more efficiently through reusing common components applications, besides being extensively researched by academics.The segment of voice interface, in turn, came up with the demand for systems capable of interacting with users, but in the application development process for this domain there is a lack of tools that make the task more productively.The FIVE (Framework for an Integrated Voice Environment) is a development environment for Voice Interface products designed to increase productivity in this segment.This paper aims to apply a SPL approach to FIVE.For this, a comparative evaluation of the process of construction of FIVE and SPL platforms was performed.Then adjustments in order to correct structural problems and, finally, the framework was validated using a set of experiments which sought to ensure the confirmation of such changes have been made.