The proliferation of malicious online behaviors, particularly cyber-trolling, presents significant challenges to maintaining healthy online communities. This paper investigates the efficacy of four deep learning architectures-BERT, LSTM, GRU, and Causal Convolutional Networks (Causal Conv 1D)-for the automatic detection of cyber-trolls based on textual content. Using a comprehensive dataset of 50,000 social media comments, we evaluate these models on their ability to distinguish between normal users and trolls. Our results indicate that while the pre-trained BERT model achieves the highest overall accuracy ($94.2 \%$), the Causal Conv 1D architecture demonstrates competitive performance $(92.7 \%)$ with significantly lower computational requirements. We also analyze the semantic features that most effectively contribute to troll detection and discuss the ethical implications of automated moderation systems. This research contributes to the development of more efficient and effective methods for maintaining civil discourse in online spaces.

Spam messages pose a significant cybersecurity threat, leading to phishing attacks, fraud, and privacy breaches. Traditional spam detection methods, such as rule-based filtering and statistical models, often fail to capture the evolving and complex nature of spam messages. In this paper, we propose an Arabic spam detection model leveraging BERT (Bidirectional Encoder Representations from Transformers), a deep learning-based NLP model. Our approach enhances classification accuracy by utilizing contextual text representations specific to the Arabic language. We preprocess Arabic text using AraBERT tokenization and fine-tune the BERT-based model on a balanced dataset of Arabic spam and ham messages. Experimental results demonstrate that our model achieves high accuracy (98%), outperforming traditional machine learning and deep learning approaches. This research highlights the potential of transformer-based models in Arabic spam filtering, paving the way for more efficient and robust detection systems.

Electrical systems keep things running in modern life, but they often run into problems like imbalances, short circuits, ground faults, and overloading, which can cause equipment to break down, fires to break out, and even large-scale blackouts. To make matters worse, acts of sabotage, physical damage, or cyberattacks on systems like SCADA can mess up operations, throw grids off balance, and set off cascading failures. To avoid these risks, there is a growing need for smarter tools that can keep track of system performance and flag potential issues before they get out of hand. In this paper, we suggest a deep learning model built on the inception architecture, designed to monitor electrical systems, call out potential security faults, and spot malicious actions. Taking advantage of deep learning, our approach helps increase fault prediction accuracy and keep operations on track.

Emotion recognition from audio signals is essential in forensic applications, offering insight into emotional states during interrogations, threat assessments, and crime scene analysis. This paper proposes an Inception-based deep learning model tailored for forensic arabic audio emotion recognition. The Inception architecture, with its multiscale feature extraction capabilities, efficiently captures subtle emotional details from complex audio signals. The model was evaluated on a dataset that represents a diverse range of emotional expressions, achieving superior performance in accuracy, robustness, and adaptability compared to traditional approaches. Its precision and ability to handle real-world variability make it particularly suited for forensic investigations. This work underscores the potential of advanced neural architectures in enhancing forensic decision-making and analysis.

In the current landscape of IT, ransomware attacks pose a major threat to cybersecurity resulting in significant monetary losses and data breaches. The detection of ransomware in time presents a challenge due to its constant evolution and complex strategies for escaping detection. This study introduces a deep learning tool —named RansFighter—based on Gated Recurrent Unit (GRU) specifically developed for ransomware detection. Our model shows, at test time, an Accuracy of 96.67%, a Precision of 97.01%, a Recall of 96.37%, an F1-Score of 96.69% and an Area Under the Curve (AUC) of 96.67%. It showcases the potential of GRUs as valuable assets to safeguard systems against ransomware threats.

CAPTCHAs, used to be an important element of security, are now facing more challenges due to the great advancements in artificial intelligence. In this paper, we investigate whether CAPTCHAs are still effective in protecting websites from automated threats. A deep learning model is suggested to automatically recognize CAPTCHA embedded characters, with performance metrics achieving an accuracy of 99.46%, an AUC of 99.98%, a precision of 99.46% and a recall of 99.43%. These findings highlight the increasing susceptibility of CAPTCHAs to sophisticated AI driven attacks and seek to emphasize the pressing importance of reevaluating CAPTCHA technologies to guarantee sustainable security.

This paper exploits the rise of artificial intelligence (AI) and deep learning (DL) to improve the use of digital forensic evidence analysis, specifically criminal identification from facial images despite disguise and makeup. Our approach leverages the VGG16 architecture for face recognition and identification, coupled with the LIME framework (Local Interpretable Model-Agnostic Explanations) to explain model recognition. This combination enables interpretation and verification of results with enhanced trust and confidence in forensic analysis. We follow a "watch and iterate" procedure, utilizing the insights generated from LIME to curate the training dataset, improving the model’s performance iteratively. The efficacy of this procedure is reflected in the remarkable outcomes: our model has an accuracy of 98.10%, precision of 98.16%, recall of 98.10%, F1-score of 98.11%, AUC of 100%. This development in forensic technology has great potential to enhance the precision and speed of criminal identification, thus leading to safer and fairer societies.

Workplace stress, a widespread issue in modern professional environments, significantly increases the potential for errors and accidents. Timely and precise stress identification is vital for fostering a secure and efficient work environment. This research introduces an innovative, comparative-analysis framework designed for real-time stress detection, utilizing Heart Rate Variability (HRV) as a reliable physiological indicator. Unlike standard heart rate measurements, HRV offers a granular view of the autonomic nervous system (ANS) function, enabling accurate stress evaluation. We implement a comprehensive methodology incorporating a refined preprocessing stage—including the removal of outliers, feature selection, and data normalization—along with a comparative assessment of eight deep recurrent neural network (RNN) architectures. These include vanilla RNN, bidirectional RNN (BiRNN), Gated Recurrent Unit (GRU), bidirectional GRU (BiGRU), standard Long Short-Term Memory network (LSTM), bidirectional LSTM (BiLSTM), Peephole LSTM, and Attention-based LSTM, applied to binary stress classification. Utilizing a dataset of 410,322 HRV records from the SWELL Knowledge Work (SWELL-KW) Dataset, our framework demonstrates exceptional performance, with the BiGRU architecture achieving a test accuracy of 99.51%. This study highlights the effectiveness of advanced temporal modeling and comparative analysis in creating robust stress detection systems for various occupational contexts, thereby enhancing workplace safety.

Intentionally mutilated fingerprints pose a significant challenge in forensic identification. Such deliberate actions typically stem from individuals seeking to evade detection or association with past or prospective criminal activities. The detection of damaged fingerprints presents a formidable obstacle for most of current forensic systems, often leading to a pronounced incidence of false negatives. The ramifications of a false negative are profound, as they preclude the establishment of links between suspects and crime scenes, impeding the acquisition of vital evidence and potentially stalling investigative progress. In response to this critical issue, this paper delves into the development of a deep learning based model expressly designed to accurately discern and capture patterns present in damaged fingerprints.

In forensic investigations, facial recognition techniques serve as critical tools for identifying and apprehending suspects. In this study, we investigate the impact of Convolutional Neural Networks (CNNs) parallelization on the performance of facial recognition models within forensic contexts. Through experiments, we demonstrate the potential benefits of parallelization in enhancing model accuracy and robustness. Leveraging a reduced dataset, we employ augmentation techniques to expand the diversity of training samples. Our findings highlight the advantages of CNN parallelization in achieving superior recognition outcomes. Nevertheless, we identify constraints linked to excessive parallelization, which may induce model overfitting.

The pervasive issue of online sexism continues to pose significant challenges, fostering environments characterized by toxicity and perpetuating harmful societal norms. In response, this paper presents an approach for the discovery of sexist statements employing convolutional neural networks (CNNs), Word Embeddings, and data augmentation techniques. Through the fusion of CNNs’ capacity for hierarchical feature extraction with the semantic representations afforded by Word Embeddings, our method achieves exemplary discrimination performance. Additionally, the incorporation of data augmentation enriches the training dataset, thereby augmenting model generalization and resilience. Empirical evaluation on a larger dataset of statements demonstrates the efficacy of our approach, surpassing many baseline approaches in terms of discovery accuracy, precision, recall and F1-score.

Handwritten signatures hold paramount importance in legal, financial, and administrative domains, necessitating the development of robust signature recognition tools for forensic applications. This paper introduces a handwritten signature recognition (HSR) model employing Parallel Convolutional Neural Networks (CNN) tailored for forensic endeavors. Utilizing the parallel processing capabilities of CNN, our proposed approach adeptly analyzes and extracts discriminative features from handwritten signature images to facilitate precise recognition. In addition, we leverage several transfer learning techniques by parallelizing proven pre-trained CNNs. Extensive experimentation validates the efficacy of our approach on a standard dataset, demonstrating high accuracy and resilience in signature recognition tasks. The proposed approach exhibits substantial promise in augmenting forensic investigations by automating signature verification processes, thereby bolstering fraud detection efforts and upholding the integrity of legal documentation.

High-performance Machine Learning (ML) models are indispensable in cybersecurity due to the need for real-time threat detection, scalability in handling large datasets, and the ability to recognize complex patterns and evolving threats. These models should reduce false positives and negatives, adapt to dynamic environments, and enable automated response mechanisms. This paper introduces an innovative methodology aimed at improving the performance and interpretability of ML models in binary classification, with a distinct emphasis on network intrusion detection. The proposed approach centers on an in-depth analysis of the confusion matrix, utilizing its insights to enhance model performance. We test our methodology on the UNSW-NB15 network intrusion dataset. We managed to improve almost all metrics with an increase in accuracy from 81.92% to 89.6%, recall from 76.42% to 89.22%, and F1 score from 82.51% to 89.76%, with the potential to obtain more improvements.

This paper introduces a novel ensemble learning methodology that leverages the confusion matrix and statistical metrics, particularly the mean, to intelligently select the most suitable model from a pool of candidates for predictive tasks on test inputs. The approach has been tested on a well-known Network Intrusion Detection Dataset (UNSW-NB15) and compared against its underlying models and other ensembling methods where it yielded superior results in three performance metrics (accuracy, recall, and F1). Out approach is model-agnostic, easily adaptable, and scalable to accommodate any number of models. Additionnally, the use of interpretable metrics such as the confusion matrix and mean enhances explainability, providing a comprehensive understanding of the model’s decision-making processes and improvement directions.

In the realm of forensic science, precise identification of individuals holds paramount importance in both investigative procedures and legal proceedings. Hands and palms recognition has emerged as a valuable biometric modality within forensic applications, owing to the distinct and intricate features inherent to these anatomical regions. The elaborate patterns of veins, creases, and ridges present on palms and fingers serve as rich sources of biometric data, crucial for accurate identification purposes. Furthermore, given the frequent involvement of hands and palms in criminal activities such as theft and assault, their recognition becomes imperative for establishing links between suspects and crime scenes. However, developing robust recognition systems tailored for forensic applications poses notable challenges, including variations in hand poses, lighting conditions, and image quality. To address these hurdles, sophisticated deep learning techniques, notably transfer learning, have been employed. By harnessing pre-trained deep learning models namely NasNetLarge, NasNetMobile, and EfficientNet, initially trained on expansive datasets for general image recognition tasks, we can adapt these models to the specific task of hands and palms recognition in forensic contexts. Our findings reveal that all three models consistently achieved over 92% accuracy across all metrics evaluated, demonstrating their efficacy as strong contenders for the hands-and-palms recognition task. Notably, the EfficientNet model exhibited superior performance compared to its counterparts, boasting more than 95.8% accuracy, precision, F1-score and recall, along with more than 98.6% specificity and 99.4% AUC.

Cyberbullying, marked by its persistent and intentional aggression online, yields severe repercussions for its victims, extending beyond immediate distress to long-lasting effects such as heightened anxiety, depression, and social withdrawal. Individuals subjected to Cyberbullying often grapple with diminished self-esteem, compromised academic performance, and strained interpersonal relations. Given the escalating prevalence of this digital menace, there is a pressing need for advanced methodologies to address it effectively. This paper introduces an approach to Cyberbullying detection, integrating techniques such as Bag-of-Words (BoW) and Term Frequency-Inverse Document Frequency (TF-IDF) analyses, along with the parallel processing capabilities of Convolutional Neural Networks (CNNs) and the contextual comprehension provided by Bidirectional Long Short-Term Memory (BiLSTM) networks. Through an experimentation on the latest Ejaz-Choudhury-Razi Cyberbullying dataset, our framework exhibits satisfactory performance in identifying instances of online hostility. These results underscore the potential of our approach to significantly contribute to ongoing efforts aimed at combating Cyberbullying in digital environments.

The Tamazight civilization stands as a significant cultural entity, marked by its linguistic diversity, historical legacy, and scriptural traditions, which collectively enrich the cultural tapestry of North Africa. Among these traditions, the Tamazight handwritten script assumes particular importance, embodying centuries of cultural identity and artistic expression. Recognizing the imperative of safeguarding this cultural heritage, our study focuses on Tamazight handwritten character recognition. Leveraging the strategic application of Transfer Learning, we explore its efficacy in this domain. Transfer Learning presents a robust framework wherein pre-existing models are adapted for specific tasks despite limited data availability. Our research employs three prominent Transfer Learning architectures: VGG, ResNet, and MobileNet. Through a rigorous comparative analysis, we discern the efficacy of these methodologies in the context of Tamazight handwritten character recognition. Our findings underscore the potential of Transfer Learning to significantly augment the accuracy and efficiency of script recognition systems, thereby advancing the overarching objective of preserving and propagating the Tamazight cultural heritage.

This paper investigates the application of GradCAM, an explainable AI (XAI) technique, to enhance the transparency and precision of fingerprint authentication systems in forensics, particularly in detecting fingerprint mutilation—a common method used to evade biometric security measures. Employing the SOCOfing dataset, which contains both unaltered and synthetically altered fingerprint images, we apply GradCAM to visualize and understand the decision-making process of a convolutional neural network (CNN) model trained to recognize and classify these alterations. Our study not only demonstrates the model’s effectiveness in identifying different types of fingerprint modifications but also identifies areas where the model’s performance can be enhanced. Through detailed visual analysis, we uncover the model’s focus points and assess its reliability across various alteration types and difficulty levels. The insights gained underline the potential of XAI in improving the robustness and reliability of biometric verification systems, paving the way for more secure and equitable AI applications in high-stakes environments.

Spam consists of unwanted messages that are often containers of malicious code and/or links pointing to shady sites or objects that pose real dangers to a company’s machines, software, or data. Spam detection is therefore a primary security objective. Nevertheless, the detection tools available on the market are few in number and their efficiency is often limited. In this paper, we propose a spam detection tool based on deep-learning. Our tool uses bidirectional Long-Short Term Memory networks while relying on Stanford Global Vectors for word representation. We present the techniques we use. Then, we conduct a series of experiments on a family of candidate detectors. Finally, we present the performance of the selected detector.

Propaganda, defamation, abuse, insults, disinformation and fake news are not new phenomena and have been around for several decades. However, with the advent of the Internet and social networks, their magnitude has increased and the damage caused to individuals and corporate entities is becoming increasingly greater, even irreparable. In this paper, we tackle the detection of text-based cyberpropaganda using Machine Learning and NLP techniques. We use the eXtreme Gradient Boosting (XGBoost) algorithm for learning and detection, in tandem with Bag-of-Words (BoW) and Term Frequency-Inverse Document Frequency (TF-IDF) for text vectorization. We highlight the contribution of gradient boosting and regularization mechanisms in the performance of the explored model.

Racism is an unequal treatment based on race, color, origin, ethnicity or religion. It is often associated with rejection, inequality, and value judgment. A racist act, whether conscious or unconscious, goes beyond insult and aggression and leaves a devastating psychological effect on the victim. Although almost all laws around the world punish racist acts and speech, racist messages are on the rise on social networks. As a result, there is a strong need for reliable and accurate detectors of racist comments to identify the offenders and take appropriate punitive action against them. In this paper, we propose a model for the detection of racist statements in text messages by Bidirectional Gated Recurrent Units. For the word representation, we use different word embedding techniques, namely Word2Vec and GloVe. We show that this combination works well and provides a good level of detection. At the end of our study, we will suggest new horizons to improve the quality of our model.

With the wide variety of applications offered by Android, this system has been able to dominate the smartphone market. These applications provide all kinds of features and services that have become highly requested and welcomed by users. Besides, these applications represent risky vehicles for malware on Android devices. In this paper, we propose a novel formal technique to enforce the security of Android applications. We start off with an untrusted Android application and a security policy, and we end up in a new version of the application that behaves according to the policy. To ensure the correctness of results, we use formal methods in each step of the process, either in the system and the security policy specification or in the enforcement technique itself. The target application is reverse-engineered to its assembly-like code, Smali. An executable semantics called k-Smali was defined for this code using a language definitional framework, called k Framework. Security policies are specified in LTL-logic. The enforcement step consists of integrating the LTL formula in the k-Smali program using rewriting. It aims to rewrite the system specification automatically so that it satisfies the requested formula.

Detecting security vulnerabilities in existing applications is a hard task. Tools to accomplish this task are not only rare but often proprietary, expensive, and not always efficient. Moreover, many of the existing tools fail to discover security vulnerabilities inside applications integrating cryptographic functionalities, since it is difficult for the inspecting software to surmount the barriers of cryptographic keys, primitives and algorithms. It is particularly tedious to cope with cryptographic protocols that may be implemented inside the inspected application. In this paper, we introduce a new tool—SinCRY—designed to inspect Java applications that implement cryptographic protocols and modules. First, we present this tool. Then, we carry out a full inspection of a legacy-like test application using this tool along with SinJAR, another static tool for inspecting Java applications through their Jar files.

The paper focuses on the conceptual development of an innovative method for authenticating sensors connected using Modbus TCP. Modbus TCP is a highly used industrial networking protocol that has very good performances but lacks security. In the first section of the paper, the authors try to highlight the importance of security of industrial control systems. The second section is dedicated to presenting the authentication process and the state-of-the-art regarding Modbus devices authentication. The third section focuses on presenting the characteristics of Modbus TCP protocol. The last section discusses the proposed conceptual solution showing a possible method of implementation.

In several critical military missions, more than one decision level are involved. These decision levels are often independent and distributed, and sensitive pieces of information making up the military mission must be kept hidden from one level to another even if all of the decision levels cooperate to accomplish the same task. Usually, a mission is negotiated through insecure networks such as the Internet using cryptographic protocols. In such protocols, few security properties have to be ensured. However, designing a secure cryptographic protocol that ensures several properties at once is a very challenging task. In this paper, we propose a new secure protocol for multipart military missions that involve two independent and distributed decision levels having different security levels. We show that it ensures the secrecy, authentication, and non-repudiation properties. In addition, we show that it resists against man-in-the-middle attacks.

Proving that a cryptographic protocol is correct for secrecy is a hard task. One of the strongest strategies to reach this goal is to show that it is increasing, which means that the security level of every single atomic message exchanged in the protocol, safely evaluated, never deceases. Recently, two families of functions have been proposed to measure the security level of atomic messages. The first one is the family of interpretation-functions. The second is the family of witness-functions. In this paper, we show that the witness-functions are more efficient than interpretation-functions. We give a detailed analysis of an ad-hoc protocol on which the witness-functions succeed in proving its correctness for secrecy while the interpretation-functions fail to do so.

In this paper, we implement hybrid Woodward-Lawson Neural Networks and weighted Fourier method to synthesize antenna arrays. The neural networks (NN) is applied here to simplify the modeling of MIMO antenna arrays by assessing phases. The main problem is obviously to find optimal weights of the linear antenna array elements giving radiation pattern with minimum sidelobe level (SLL) and hence ameliorating the antenna array performance. To attain this purpose, an antenna array for reliable Multiple-Input Multiple-Output (MIMO) applications with frequency at 2.45 GHz is implemented. To validate the suggested method, many examples of uniformly excited array patterns with the main beam are put in the direction of the useful signal. The Woodward-Lawson Neural Networks synthesis method permits to find out interesting analytical equations for the synthesis of an antenna array and highlights the flexibility between the system parameters in input and those in output. The performance of this hybrid optimization underlines how well the system is suitable for a wireless communication and how it participates in reducing interference, as well.

In this paper, we are proposing a new tool for reversing Java applications called SinJAR. SinJAR is a lightweight software written in Java aiming at inspecting bytecode at compile time and producing the structure tree of a targeted application. Besides, it is able to detect vulnerabilities and security weaknesses inside the Java code. SinJAR can be used for two purposes. The first one is sane and consists in using it to verify whether or not an application is safe and compliant with its specification. The second one is malicious and consists in spying applications through their bytecode and exploiting vulnerabilities that they may enclose. In this paper, we will show how to detect SinJAR malicious actions after showing the capabilities of the tool through few ad hoc attack scenarios conducted in a real military context.

In this paper, we are proposing a new formal framework for reasoning on authentication in security protocols based on analytic functions. We give sufficient conditions that, if satisfied, the protocol is declared correct with respect to authentication. We validate our approach on the Yahalom-Lowe protocol. First, we show that it satisfies these few conditions, thus, we conclude that it is correct for authentication.

The concepts of array processing and smart antenna give a promising solution to the significant increase of data rates in wireless transmission systems. In this paper, we deal with the problem of designing linear antenna arrays for specific radiation properties of MIMO applications based on Direction-Of-Arrival estimation and Taylor beamforming techniques. The objectives of this paper can be summarized as to minimize the maximum sidelobe level (SLL), combined the Taylor method and MUSIC (Multiple Signal Classification) algorithm. The performance of this hybrid optimization determines how well the system is convenient for a reliable wireless communication and interference reduction. This paper will discuss the application of MUSIC algorithm for linear array antenna (4, 8 and 16 antennas) in order to estimate the Direction-Of-Arrival of various angles of elevation and azimuth.

In this paper, we use witness-function to capture attack scenarios in cryptographic protocols. A witness-function is a protocol-dependent metric that attributes a reliable security level to every atomic message. We use these functions to prove the protocol correctness with respect to secrecy by proving that the security level of every atomic message never decreases throughout all consecutive receiving and sending steps of the protocol. In this paper, we analyze the defective variant of the Otway-Rees protocol and we demonstrate that the use of witness-functions can be a key element in tracing a well-known type flaw that this protocol involves.