Predicting network metrics with high accuracy is a challenging task. Network Latency prediction allows Network Operators (NO), Internet Service Providers (ISP), and Over The Top (OTT) Service Providers to optimize their performance in almost real-time. Moreover, applications can make decisions based on latency predictions, improving the quality of the provided services. In recent years, many works proposed machine learning based techniques to predict network metrics, especially using Recurrent Neural Network (RNN) techniques, such as Long Short-Term Memory (LSTM) Networks. Nevertheless, despite the good results achieved, the computational cost of training and keeping a model updated makes adopting those techniques unfeasible in some scenarios. In this work, we propose the usage of Data Stream Learning techniques to predict RoundTrip Time (RTT) and One-Way Delay (OWD) metrics using realworld data. The experiment results demonstrated a prediction performance similar to LSTM networks using only a fraction of the computational resources used by LSTM.

Abstract The number of malicious software applications, or malware programs, increases every year. Their development becomes more sophisticated as new techniques are used to bypass program scanning software applications, such as antiviruses. Thereby, deep learning‐based methods emerge as a new promising way to identify these threats. Our main purpose and contribution in this work is proposing and implementing a successful approach to tackle both binary and multiclass malware classification problems. We used unsupervised word embedding algorithms for representing software applications to be analyzed and long‐short term memory for classifying the software applications. For evaluating our pipeline, we introduce a new dataset for binary and multiclass malware classification because we could not find large datasets containing sufficient samples of cleanware and the various malware types for multiclass classification that could be used to evaluate classification models. Our experimental results reached an accuracy of 88.94% for binary classification and 75.13% for multiclass classification. These results suggest that the proposed dataset is challenging, and using it can help in the training of better malware classifiers, improving security.

The use of Crowdsourcing to solve public problems is called Citizen-Sourcing and shows the potential to increase citizen participation in the context of e-government. Successful implementations of Citizen-Sourcing applications require the citizen to continuously engage with each other and with the government through these applications. In general, citizens expect, among other things, that the government responds to their comments in an application by immediately solving the problems pointed out or indicating when and how they would be solved. In a literature review, we could not find any model for adequately assessing the quality of the service provided by local e-governments for citizens in Citizen-Sourcing scenarios. Hence, we propose an approach to analyze the quality of the government's response to the citizen through Citizen-Sourcing applications. To validate the proposed approach, we have conducted a case study using real data collected from Colab.re, a Citizen-Sourcing application very popular in Brazil. Our main contributions are a novel approach to assess the quality of interaction between local e-government and citizen and a detailed discussion regarding a local e-government's action in the platform.

In higher education, many students exceed the expected time to complete their undergraduate programs. This delay is called retention, which can lead to program abandonment. STEM undergraduate programs, in particular, have higher retention and dropout rates when compared to other non-STEM programs. The students in such programs end up exchanging or dropping out the programs before graduating, causing waste in economic, social and academic terms. In this context, Recommendation Systems can be used to support students and managers in choosing disciplines, contributing for them achieving better academic performance and thus aiming to improve student learning and engagement and mitigate retention. For constructing these Recommendation Systems, Educational Data Mining techniques, including machine learning algorithms, can be used to identify and predict retention situations and contribute to reducing their occurrence. The aim of this paper is to present a Systematic Literature Review (SLR) for identifying the use of Educational Data Mining methodologies, techniques and tools to implement Recommendation Systems with a focus on preventing student retention in higher education programs. We selected studies available in digital libraries that are international references in publications of scientific articles, in order to answer the following research question: What machine learning methods were used in Recommendation Systems in the context of Educational Data Mining? Among the various studies found to reduce student retention rates, most used methods to predict student grades. We observed that there are many papers proposing the use of machine learning methods for predicting failure in disciplines, either through regressors or classifiers. However, just a few studies have proposed Recommendation Systems to assist students in choosing subjects at the time of enrollment for the next term, which indicates a large area for the development of further future work in this field.

This Research to Practice Full Paper presents a pedagogical practice for technical high school students for developing Computational Thinking (CT) abilities through Internet of Things technologies. The covered topics in our proposal include (i) the use of temperature and humidity sensors for data collection, treatment and visualization using Arduino and micro-controllers; (ii) Smart and Human Cities (SHC) and Open Data concepts, in order to lead the students to reflect on the problems of their city and on data protection. Our motivation to our proposal is due to the educational processes have to develop criticality and the ability to solve problems among students. In this context, CT has been used for this through the use of robotics, game building or unplugged computing. On the other hand, technologies for implementing Internet of Things (IoT) have been used in several domains of society, including cities transformation. One important aspect in this scenario is data generation, which have to be carefully tackled by government's and who develop solutions to SHC. In this way, using IoT for teaching CT is an important aspect, also considering open data, privacy and SHC context. In our pedagogical practice, students were able to design and develop solutions to problems in their daily lives indirectly applying CT skills, such as decomposition, pattern recognition, abstraction, automation and analysis, as well as self-skills, collaboration, creativity and critically, required nowadays in broad professional training. It was also possible to develop students' interest in raising awareness of the use of computational technologies, as a solution to problems in society considering aspects of SHC and open data; propose a technological solution using IoT; and analyze the use of these data collected for the social well-being. For the evaluation of our proposal, we carried out questionnaires and tasks observation. The experience was considered successful in its planning and application, with a positive evaluation of the participating students.

Identifying malware has always been a great challenge. Much money and time has been invested by companies and governments to mitigate the impact of these threats. Nowadays, with the increasing amount of data available, it is possible to use more precise classification techniques. However, most large datasets that include malicious and non-malicious softwares are not public, which hinders the quest for solutions based in technologies that rely on the availability of large amounts of data, such as deep learning. To overcome this limitation, this article introduces a new large dataset for malware classification, which was made publicly available. We then propose a model to train a multiclass classification recurrent neural network (RNN), more specifically a long short-term memory neural network (LSTM) on our dataset. This model for analyzing unstructured malware data is then tested on unseen programs and the accuracy obtained reaches 67.60%, including six classes with five different types of malware.

The use of the latent factor models technique overcomes two major problems of most collaborative filtering approaches: scalability and sparseness of the user's profile matrix.The most successful realizations of latent factor models are based on matrix factorization.Among the algorithms for matrix factorization, alternating least squares (ALS) stands out due to its easily parallelizable computations.In this work we propose a methodology for comparing the performance of two parallel implementations of the ALS algorithm, one executed with MapReduce in Apache Hadoop framework and another executed in Apache Spark framework.We performed experiments to evaluate the accuracy of generated recommendations and the execution time of both algorithms, using publicly available datasets with different sizes and from different recommendation domains.Experimental results show that running the recommendation algorithm on Spark framework is in fact more efficient, once it provides in-memory processing, in contrast to Hadoop's twostage disk-based MapReduce paradigm.

Information and Communication Technologies are indispensable components of smart cities. Its applications are present in several areas, such as urban mobility, environmental issues and medical systems. In this scenario, the use of crowdsourcing technologies comes to help people to contribute to the development and improvement of the urban digital services. However, using crowdsourced data in smart cities solutions can lead to problems with the security and the privacy of user's data. The setting of comprehensive Functional Requirements (FR) to ensure data privacy is an approach for preventing the occurrence of such issues. In this work, we intend to identify, from a literature review the main privacy requirements that have been observed in the development of applications that make use of crowdsourced data in Smart Cities scenarios.

Visualizing data in tables usually is not the best way to help users understanding large amounts of data. Instead, data visualization in graphical format, such as charts, has been used to explore available data. A good data visualization is defined as a well-designed presentation of interesting data, aiming to communicate ideas with clarity, precision and efficiency. Nowadays, considering the open data movement, many open data portals offer different types of data visualization. However, when browsing some of these portals, we can find many bad data visualizations, frequently ambiguous, confusing and unusable. Hence, it is essential to have an instrument capable of analyzing the quality of data visualizations, helping the designer to use the full capacity of a data visualization to provide a more efficient resource information to the users. This work aims to present an instrument that integrate a set of heuristics to assess the quality of data visualizations. Such heuristics were chosen due to have been previously proposed in several works in literature, and proved successful. To evaluate the proposed instrument, we conducted an experiment with a group composed by computer science graduate students. We analyzed the results using Cohen's Kappa and Any-two agreement statistical tests, which indicated that the instrument is adequate.

For a given data set, different learning algorithms typically provide different classifiers. Although it is possible to simply select the most successful classifier, the less successful classifiers could have potentially valuable information that may be wasted. This work proposes GAESC, an algorithm for evolving a set of classifiers into a single symbolic classifier using genetic algorithms. Individuals are formed by rules collected from symbolic classifiers and rules from association classification rules. Experimental results in three data sets from UCI show that GAESC outperforms the single symbolic classifiers in terms of classification error rate.

Learning algorithms are an integral part of the data mining (DM) process. However, DM deals with a large amount of data and most learning algorithms do not operate in massive datasets. A technique often used to ease this problem is related to data sampling and the construction of ensembles of classifiers. Several methods to construct such ensembles have been proposed. However, these methods often lack an explanation facility. This paper proposes methods to construct ensembles of symbolic classifiers. These ensembles can be further explored in order to explain their decisions to the user. These methods were implemented in the ELE system, also described in this work. Experimental results in two out of three datasets show improvement over all base-classifiers. Moreover, according to the obtained results, methods based on single rule classification might be used to improve the explanation facility of ensembles.