In this paper, we propose a Machine Learning-based approach to validate suggested learning materials. Learning material validation is an essential part of the learning process, ensuring that learners have access to relevant and accurate information. However, the process of manual validation can be time-consuming and may not be scalable. Traditional learning contents are often only updated or changed in the yearly course revisions. This can be presented with some challenges, especially to courses on emerging subjects and catering to diversified learners, which includes the ability to provide adaptive and updated learning contents to the learners, and the opportunity to continually incorporate feedback. We present a solution and framework that utilizes machine learning algorithms to validate learning materials in an open learning content creation platform. Our approach involves pre-processing the data using Natural Language Processing techniques, creating vectors using TF-IDF and training a Machine Learning model to classify the subject of the learning material. We then calculate the similarity with existing materials for the given course to make sure there is not an existing mate-rial with same content and the new material will add new value. Using an augmented TF-IDF score, we check if the suggested learning materials satisfies the key phrases for the course. We evaluate our approach by comparing the Machine-Learning based approach to manual validation. Not only does the machine-learning based approach reduce the time and effort needed for validation, but it also achieves high accuracy in detecting duplicates and similarity matches.
The recent increase in the frequency and diversity of cyber-attacks has raised concerns among companies, organizations, and users. Web applications are particularly critical among the various targets as they provide users access to online data and financial transactions. SQL injections, which can be exploited to compromise the security of web infrastructures, represent a significant risk. While SQL injection detection has been a solved problem for scenarios where all network layer datagrams are analyzed, it remains a challenge in data-intensive networks that use lightweight protocols based on network flows, such as NetFlow. In this paper, we attempt to emulate a realistic SQL injection attack scenario, where an attacker tries to generate minimum noise in the network. To this end, we generated SQL injection attack datasets based on flows using the SQLMap tool and the keep-alive parameter. We evaluated several machine learning algorithms and achieved a detection rate higher than 97
In this work, a systematic literature review of methods used for SQL injection detection based on intelligent algorithms, especially deep learning or machine learning, is carried out. We identify the main methods used, the algorithms that have been tested and the main techniques used to process the data. In order to make the study, a quantitative review has been performed following the SLR methodology which includes identification of the need for the review, definition of the search strategy, search and selection of articles, and data extraction and analysis. This work aims to prove if it is possible to apply intelligent algorithms in this field and to collect the methods used so far.
The high prevalence rates of obesity and its comorbidity with serious diseases like diabetes or hypertension and psychological problems like depression or anxiety, highlight the necessity of new treatments for this condition. This is a pilot study, which main objective is analyzing the correct functioning of the program with a small sample. The sample of this pilot study is composed by 27 participants. The En_Línea program to change lifestyle in people with overweight and obesity is based on the LEARN program. The results showed significant differences between pre-intervention weight and the middle treatment assessment at week 8 (n = 9, z = −2.521, p = 0.012, r = 0.84). Significant differences were also observed between pre- and post-intervention weight (n = 6, z = −2.201, p = 0.028, r = 0.9) and at 3 months follow-up (n = 5, z = -2.023, p = 0.043, r = 0.9). These results were supported by high effect sizes. In general, the En_Línea program seems a tool with a great potential but presents some limitations, mainly the low adherence rates or the length of the program. This pilot study was helpful to identify and address these limitations before its application with a larger sample.
This paper analyzes the robustness and stability of a published methodology to improve the evaluation of complex projects in university courses. For this purpose, different types of experiments are performed on a dataset (e.g. elimination of features, input perturbations) of a subject in Computer Systems at the University of La Rioja (Spain); then, the methodology is reapplied, analyzing whether the final conclusions remain similar. The results show that the conclusions obtained, despite the variations introduced, are consistent.
A system of stochastic differential equations is proposed as a model to describe the diffusion of malware in a wireless sensor network. The system is obtained from a SEIR model by perturbing the rate at which susceptible sensors become exposed using a Gaussian white noise. Then, the stochastic stability of the model is analysed to obtain a sufficient parameter relation for the asymptotic stability in probability of the equilibrium.
The use of Machine Learning techniques and technological resources that facilitate diagnosis and intervention at early ages (0–6 years) will facilitate the development of both processes from the point of view of accuracy. This paper analyses the most useful Machine Learning techniques to be applied to diagnosis and therapeutic intervention in the field of early care. It also describes the development of a web application, eEarlyCare, which includes the recording and interpretation of the results through Learning Analytics techniques of the observation of different early development problems. In addition, an evaluation of the usability of this web application is carried out as part of a training project aimed at updating the technological and data analysis strategies of early intervention professionals. The results support the use of this type of computer applications in which learning analytics and visualisation of results techniques are included. The proposals for improvement focus on the use of technological resources similar to intelligent voice assistants that regulate the work of the therapy professional. Further studies will address these proposals for improvement.
Many models have appeared in the scientific literature with the aim of simulating the spread of malware on wireless sensor networks. Usually, these are global and unrealistic models that do not properly consider the characteristics of malware specimens and networks. In this paper a small critical analysis of this fact is carried out using the MDBCA model as an example. This study is illustrated with an improved model proposal.
In recent years, the proliferation of Internet of Things (IoT) devices has led to an increase in network DDoS attacks. This requires effective methods to classify traffic on IoT networks as benign or DDoS-vulnerable. The present research compares the performance of some of the existing classification models to determine which is better at detecting suspicious packets from an IoT network. We evaluate six different Machine Learning (ML) models: linear, instance-based, SVM, probabilistic, tree-based, and boosted ones. The analyzed dataset contains labeled traffic packets, used to train and test the models on real-life data. Model performance is benchmarked in terms of the standard metrics: accuracy, precision, recall, TN Ratio, and F1-score. The outcomes of this study can be applied to improve IoT network security in Industry 4.0 environments as it provides valuable insights into the most effective ML algorithms to classify IoT network traffic.
Explainable artificial intelligence aims to describe an artificial intelligence model and its predictions. In this research work, this technique is applied to a subject of a Computer Science degree where the programming language changed from Octave to Python. Experiments are performed to analyze the explainability using the SHapley Additive exPlanations algorithm for XGBoost regressor model (for numerical grade prediction) and XGBoost classifier model (for class grade prediction). After the validation and training process, several conclusions are drawn that validate the idea of changing the programming language to a more popular one such as Python. For example, regarding classification problems, the most important feature for the insufficient class in the Octave courses is the practical exam.
In the Master’s course in Energy, taught at the Faculty of Physics of the Complutense University of Madrid, the optional subject “Modeling and Simulation of Energy Systems. Projects” is part of the courses, among other subjects. This course has a practical nature, which facilitates the dynamisation of the class. The learning-by-doing approach is proving very useful for students to acquire new concepts and learn to use software tools they are not familiar with. This paper describes a series of actions that have been carried out with the aim of making classes more dynamic and facilitating learning. The range of activities has been very varied, from the presentation of articles, the discussion of practical cases, computer practices, team work, ... The results both in grades and in the degree of satisfaction of students and teachers are very good.
Modern technical universities help students get practical experience. They educate thousands of students and it is hard for them to connect individual students with relevant industry experts and opportunities. This article aims to solve this problem by designing a matchmaking procedure powered by a recommendation system, an ontology, and knowledge graphs. We suggest improving recommendations and reducing the cold-start problem with a re-ranking module based on student educational profiles for students who opt-in. Each student profile is represented as a knowledge graph derived from the successfully completed courses of the individual. The system was tested in an online experiment and demonstrated that recommendations based on student educational profiles and their interaction history significantly improve conversion rates over non-personalised offers.
One of the critical challenges for natural language processing methods is the issue of automatic content summarization. The enormous increase in the amount of data delivered to users by news services leads to an overload of information without meaningful content. There is a need to generate an automatic text summary that contains as much essential information as possible while keeping the resulting text smooth and concise. Methods of automatic content summarization fall into two categories: extractive and abstractive. This work converts the task of extractive summarization to a binary classification problem. The research focused on analyzing various techniques for extracting the abstract in a supervised learning manner. The results suggest that this different view of text summarization has excellent potential.
We often come across the seemingly obvious remark that the modern world is full of data. From the perspective of a regular Internet user, we perceive this as an abundance of content that we unintentionally consume every day, including links and amusing images that we receive from friends and content providers via webpages, social media, and other sources. Consequently, some of this information is only loosely related to the truth. This problem is one of the challenges the SWAROG project is intended to address. SWAROG is an ongoing Polish research project, which involves the creation of artificial intelligence algorithms for the automatic classification and detection of so-called fake news. In this paper, we report the recent project’s achievements regarding fake news detection, analyse and discuss the pitfalls the existing solutions run into concerning data annotation, and explain the project approach to deliver services for determining the credibility of information published in public space.
Educational methods have changed significantly in recent years. In this sense, the COVID-19 pandemic has led to an increase of remote teaching. Specifically, laboratory practices with real systems are essential in the field of engineering. The lack of physical plants and the implementation of Blended Learning experiences force the development of emulated laboratory plants. This work proposes a new approach with the identification and implementation of a specific section of an enhanced level control plant located at the Polytechnic Engineering School of Ferrol, using a low-cost embedded system. In this case, the solution is equipped with an improved design of a synoptic visualization of the plant developed with a Node-RED application. This emulation allows students to attend practice lessons of control subjects remotely by means of flexible hardware and software tools.
Advanced persistent threats are cyberattacks characterized by its complexity, persistence and stealth. One of the basic tools employed in an APT campaign is specific specimens of advanced malware whose malicious payload consists of infecting some concrete devices. Consequently, this type of malware needs to have some type of knowledge of the network and devices. The main goal of this work is to introduce a novel model to obtain the most efficient path that a malware must follow to achieve its objective when no kind of information about the devices and network is known. The proposed model is based on Q-Learning methodology and it allows to consider some security countermeasures like honeypots (the model is able of find a path that avoids these honeypots). Furthermore, in order to avoid that APT malware gathers the information of the network, we propose using Moving Target Defense (MTD) which does not avoid malware propagation but it triggers that malware learns in a not proper way.
Teaching Artificial Intelligence in higher education develops critical thinking, problem-solving, and computational skills. Negotiation is a crucial aspect of multi-agent systems, enabling agents to achieve their goals through communication and collaboration. In this area, simulation platforms provide a flexible and safe way to experiment, leading to improvements in negotiation and decision-making in a wide range of scenarios. Board games, which include interaction and negotiation between players, provide a low-cost and low-risk way to experiment with different negotiation strategies. In this context, we present a novel simulation platform based on the rules of Catan, a popular boardgame that entails both strategic thinking and negotiation. This platform is oriented to teach negotiation in artificial intelligence and multi-agent systems. The platform allows students to develop and program intelligent agents to play autonomously, improving their technical skills and applying their understanding of Artificial Intelligence concepts.
Usually statistical techniques are employed to analyze malware behavior mainly through machine learning-based methods. However it seems legitimate to wonder if some statistical methods could be useful as a complementary tool to malicious code propagation models. This work explores this possibility with a first (and simple) application of the use of survival analysis to the study of the simulations obtained from a compartmental and individual SI model whose dynamics is described by means of a cellular automaton. The results obtained are in line with what could reasonably be expected.
Phishing attacks continue to pose a significant threat to internet security, with phishing URLs being among the most prevalent attacks. Detecting these URLs is challenging, as attackers constantly evolve their tactics. Few-shot learning has emerged as a promising approach for learning from limited data, making it ideal for the task of phishing URL detection. In this paper, we propose a prototypical network (DPN) disentangled by triplet sampling that learns disentangled URL prototypes to improve the accuracy of phishing detection with limited data. The key idea is to capture the underlying structure and characteristics of URLs, making it highly effective in detecting phishing URLs. This method involves sampling triplets of anchor, positive, and negative URLs to train the network, which encourages the embedding space to be more separable between phishing and benign URLs. To evaluate the proposed method, we have collected and assessed a real-world dataset consisting of one million URLs, and additionally utilized two benchmark URL datasets. Our method outperforms the state-of-the-art models, achieving accuracies of 98.0
Over the past decade, mobile devices have become an integral part of our daily lives. These devices rely on applications to deliver a diverse range of services and functionalities to users, such as social networks or online shopping apps. The usage of these applications has led to the emergence of novel security risks, facilitating the rapid proliferation of malicious apps. To deal with the increasing numbers of Android malware in the wild, deep learning models have emerged as promising detection systems. In this paper, we propose an Android malware detection system using Convolutional Neural Networks (CNN). To accomplish this objective, we trained three distinct models (VGG16, RESNET50, and InceptionV3) on the image representation of the Dalvik executable format. Our assessment, conducted on a dataset of more than 13000 samples, showed that all three models performed up to 99
