Mobile devices have become indispensable, existing authentication methods such as PINs and passwords verify users only once at login, leaving devices exposed afterward. Continuous Authentication (CA) mitigates this risk by continuously verifying user identity during interaction. Among CA techniques, touch-based behavioral biometrics are particularly promising because they require no special hardware and can unobtrusively capture unique user patterns. This paper proposes a hybrid one-class ensemble framework for robust continuous authentication using large-scale mobile typing data from over 30,000 users[1]. Our approach combines One-Class Support Vector Machine (OC-SVM) and Isolation Forest (IF) anomaly detectors through score-level fusion, reinforced with Hard Negative Mining (HNM) and cross-shard impostor simulation to ensure generalization against unseen attackers. Using a 12-feature standardized set capturing timing, rhythm, and error dynamics, our system achieves an average AUC of 0.97 and EER of 2.9%, with over 90% of users below 5% EER. These results demonstrate that ensemblebased one-class modeling with hard negative refinement can deliver exceptionally reliable continuous authentication without requiring labeled impostor data.

Most existing touch-based continuous authentication systems report promising accuracy under controlled conditions, but they provide limited insight into their feasibility in real-world mobile deployments. This paper evaluates mobile typing behavior as a scalable behavioral biometric while explicitly accounting for system-level deployment constraints. We use a large-scale observational dataset comprising approximately 1.65 million keystroke-level interactions from tens of thousands of users [1]–[3], collected in an uncontrolled web-based environment. In addition to standard authentication metrics, we systematically measure inference latency, memory footprint, and serialized model size for multiple modeling approaches. Our results show that while ensemble models achieve superior authentication performance, simpler or more efficient models may be preferable under strict resource constraints, enabling continuous or on-device operation. These findings demonstrate that mobile typing-based authentication can be both effective and deployable at population scale, and that realistic system feasibility considerations are essential when designing behavioral biometric authentication systems.

Mobile devices have become an integral part of daily life and ensuring secure and continuous access to them has become a critical concern. Various traditional authentication methods, such as passwords, PINs, and pattern locks, authenticate users only on entry. Although these security measures are being upgraded, they still fail to secure the devices post login. This leaves devices exposed during active sessions. Continuous Authentication provides an alternative by verifying a user's identity continuously rather than only at initial login. This study reviews and compares different ML algorithms used in various Touch Based CA schemes based on touch dynamics. It will give a comprehensive overview of how ML algorithms have been applied, highlighting trends in performance and their challenges.

Deepfake technology, driven by ML and DL, is both a creative tool and a threat to privacy, security, and trust. While it enables innovation in entertainment and education, its misuse raises ethical concerns. This paper explores deepfake detection, analyzing 33 studies (2020–2023). It compares ML-based methods, which are efficient but less accurate, with DL-based approaches that are precise but resource-intensive. Key factors include detection accuracy, real-world adaptability, and resistance to evasion tactics. Challenges include limited training data, scalability issues, and evolving deepfake techniques. This research identifies gaps and suggests improvements to ensure trust, accuracy, and usability in detection systems.

This paper explores Federated Learning (FL) for enhancing security in the Internet of Things (IoT) through intrusion detection systems (IDS). By using a decentralized approach, we focus on improving privacy and reducing data centralization. Our implementation utilizes artificial neural networks (ANNs) across multiple clients to identify and classify cybersecurity threats at their source. This method preserves data privacy and allows for local data processing, facilitating effective IoT device distribution. We evaluate FL’s potential for implementing IDS in IoT environments, using a robust ANN as the shared model and federated averaging (FedAvg) as the aggregation algorithm. Experiments conducted with the Car Hacking dataset and another dataset focus on binary and multiclass classification. Distributed devices classify data without sharing information, thus maintaining privacy. Our results indicate that the collaborative FL-based Deep Learning IDS (FL DL-IDS) yields more accurate results compared to centralized methods, making it a viable option for IoT-IDS. We also tested FedAdam and FedAdagrad aggregation algorithms, finding that FedAvg performs comparably in our scenarios.

Human Activity Recognition (HAR) has gained increasing attention due to its wide-ranging applications in daily life, healthcare, and smart systems. The ability to accurately identify and classify human activities has significant implications for health monitoring, fitness tracking, and improving user interactions with technology. While deep learning approaches are often explored, this paper focuses on comparing traditional machine learning models, including Decision Trees, K-Nearest Neighbors (KNN), Logistic Regression, Naive Bayes, and Random Forest. These models are trained using three distinct datasets containing only the raw accelerometer data. The study evaluates each model based on classification accuracy, recall and F1 Score to highlight their strengths and limitations. Results show that Random Forest and Decision Tree machine learning algorithms were the best at classifying activities, as well as common mistakes among similar activities throughout all the models used. These results set the framework for future work in wearable technology and validation of that work.

Home security is of paramount importance in today's world, where we rely more on technology, home security is crucial. Using technology to make homes safer and easier to control from anywhere is important. Home security is important for the occupant’s safety. In this paper, we came up with a low cost, AI based model home security system. The system has a user-friendly interface, allowing users to start model training and face detection with simple keyboard commands. Our goal is to introduce an innovative home security system using facial recognition technology. Unlike traditional systems, this system trains and saves images of friends and family members. The system scans this folder to recognize familiar faces and provides real-time monitoring. If an unfamiliar face is detected, it promptly sends an email alert, ensuring a proactive response to potential security threats.

Geographically-distributed wide-area networks (WANs) offer expansive distributed and parallel computing capabilities. This includes the ability to advance Wide-Area Big Data (WABD). As data streaming traverses foreign networks, intrusion detection systems (IDSs) and intrusion prevention systems (IDSs) play an important role in securing information. The authors anticipate that securing WAN network topology with IDSs/IPSs can significantly impact wide-area data streaming performance. In this paper, the researchers develop and implement a geographically distributed big data streaming application using the Python programming language to benchmark IDS/IPS placement in hub-and-spoke, custom-mesh, and full-mesh network topologies. The results of the experiments illustrate that custom-mesh WANs allow IDS/IPS placements that maximize data stream packet transfers while reducing overall WAN latency. Hub-and-spoke network topology produces the lowest combined WAN latency over competing network designs but at the cost of single points of failure within the network. IDS/IPS placement in full-mesh designs is less efficient than custom-mesh yet offers the greatest opportunity for highly available data streams. Testing is limited by specific big data systems, WAN topologies, and IDS/IPS technology.

This research aims to further understanding in the field of continuous authentication using behavioral biometrics. We are contributing a novel dataset that encompasses the gesture data of 15 users playing Minecraft with a Samsung Tablet, each for a duration of 15 minutes. Utilizing this dataset, we employed machine learning (ML) binary classifiers, being Random Forest (RF), K-Nearest Neighbors (KNN), and Support Vector Classifier (SVC), to determine the authenticity of specific user actions. Our most robust model was SVC, which achieved an average accuracy of approximately 90%, demonstrating that touch dynamics can effectively distinguish users. However, further studies are needed to make it viable option for authentication systems

In the realm of computer security, the importance of efficient and reliable user authentication methods has become increasingly critical. This paper examines the potential of mouse movement dynamics as a consistent metric for continuous authentication. By analyzing user mouse movement patterns in two contrasting gaming scenarios, "Team Fortress" and Poly Bridge we investigate the distinctive behavioral patterns inherent in high-intensity and low-intensity UI interactions. The study extends beyond conventional methodologies by employing a range of machine learning models. These models are carefully selected to assess their effectiveness in capturing and interpreting the subtleties of user behavior as reflected in their mouse movements. This multifaceted approach allows for a more nuanced and comprehensive understanding of user interaction patterns. Our findings reveal that mouse movement dynamics can serve as a reliable indicator for continuous user authentication. The diverse machine learning models employed in this study demonstrate competent performance in user verification, marking an improvement over previous methods used in this field. This research contributes to the ongoing efforts to enhance computer security and highlights the potential of leveraging user behavior, specifically mouse dynamics, in developing robust authentication systems.

In recent decades, mobile devices have evolved in potential and prevalence significantly while advancements in security have stagnated. As smartphones now hold unprecedented amounts of sensitive data, there is an increasing need to resolve this gap in security. To address this issue, researchers have experimented with biometric-based authentication methods to improve smartphone security. Following a comprehensive review, it was found that gait-based mobile authentication is under-researched compared to other behavioral biometrics. This study aims to contribute to the knowledge of biometric and gait-based authentication through the analysis of recent gait datasets and their potential with machine learning algorithms. Two recently published gait datasets were used with algorithms such as Random Forest, Decision Tree, and XGBoost to successfully differentiate users based on their respective walking features. Throughout this paper, the datasets, methodology, algorithms, experimental results, and goals for future work will be described.

Throughout the past several decades, mobile devices have evolved in capability and popularity at growing rates while improvement in security has fallen behind. As smartphones now hold mass quantities of sensitive information from millions of people around the world, addressing this gap in security is crucial. Recently, researchers have experimented with behavioral and physiological biometrics-based authentication to improve mobile device security. Continuing the previous work in this field, this study identifies popular dynamics in behavioral and physiological smartphone authentication and aims to provide a comprehensive review of their performance with various deep learning and machine learning algorithms. We found that utilizing hybrid schemes with deep learning features and deep learning/machine learning classification can improve authentication performance. Throughout this paper, the benefits, limitations, and recommendations for future work will be discussed.

A deepfake is a photo or video of a person whose image has been digitally altered or partially replaced with an image of someone else. Deepfakes have the potential to cause a variety of problems and are often used maliciously. A common usage is altering videos of prominent political figures and celebrities. These deepfakes can portray them making offensive, problematic, and/or untrue statements. Current deepfakes can be very realistic, and when used in this way, can spread panic and even influence elections and political opinions. There are many deepfake detection strategies currently in use but finding the most comprehensive and universal method is critical. So, in this survey we will address the problems of malicious deepfake creation and the lack of universal deepfake detection methods. Our objective is to survey and analyze a variety of current methods and advances in the field of deepfake detection.

Abstract— Conspicuous progression in the field of machine learning (ML) and deep learning (DL) have led the jump of highly realistic fake media, these media oftentimes referred as deepfakes. Deepfakes are fabricated media which are generated by sophisticated AI that are at times very difficult to set apart from the real media. So far, this media can be uploaded to the various social media platforms, hence advertising it to the world got easy, calling for an efficacious countermeasure. Thus, one of the optimistic counter steps against deepfake would be deepfake detection. To undertake this threat, researchers in the past have created models to detect deepfakes based on ML/DL techniques like Convolutional Neural Networks (CNN). This paper aims to explore different methodologies with an intention to achieve a cost-effective model with a higher accuracy with different types of the datasets, which is to address the generalizability of the dataset.

The aim of this undergraduate research paper is to investigate the use of continuous authentication with mobile touch dynamics, using three different algorithms: Neural Network, Extreme Gradient Boosting, and Support Vector Machine. Something that is starting to occur in the modern day is the use of mobile devices rising steadily, and in today's world, the number of mobile phone subscriptions has surpassed 6 billion [1]. What we will be looking into is the touch dynamics of users and how their interactions on a device can be utilized for authentication and security. The term touch dynamics refers to patterns from an individual, these patterns being completely unique, on a mobile device via the touch screen. These interactions could be swipes, taps, the amount of pressure on the screen, and several more. The use of this type of data expands well into continuous authentication for security. Continuous authentication refers to monitoring and verifying a user's identity on a device continuously, rather than just a login or a one-time security feature [2]. For this research to have enough data for algorithms to utilize, we collected touch data from 40 participants, since this is seen to be a good starting point [3],[4], using the LG V30+. The way this data was collected was through each participant playing 4 different mobile games, Minecraft, Slither.io, PUBG, and Diep.io for 10 minutes each game. Since these games are high-action games that utilize two-finger input, it allowed us to gather a lot more data from everyone. Once the data was collected and cleaned, the three algorithms stated before were then trained and tested, and their performance was evaluated based on their Accuracy, F1 Score, False Negative Rate, and False Positive Rate. The results from each of these models show that they were able to accurately classify unique users with an accuracy ranging from 75% to 95%, very similar to [5]. The SVC and XGBoost algorithms performed slightly behind the Neural Network, with the Neural Network achieving an accuracy of 90.04%. These results show us that touch dynamics can uniquely identify individuals based on patterns they make on a device. We also learn that the type of model that is decided for recognizing and learning these patterns matters a lot. Different algorithms have slightly different functionality which can affect the results in small ways.

Engineering more secure software has become a critical challenge in the cyber world. It is very important to develop methodologies, techniques, and tools for developing secure software. To develop secure software, software developers need to think like an attacker through mining software repositories. These aim to analyze and understand the data repositories related to software development. The main goal is to use these software repositories to support the decision-making process of software development. There are different vulnerability databases like Common Weakness Enumeration (CWE), Common Vulnerabilities and Exposures database (CVE), and CAPEC. We utilized a database called MITRE. MITRE ATT&CK tactics and techniques have been used in various ways and methods, but tools for utilizing these tactics and techniques in the early stages of the software development life cycle (SDLC) are lacking. In this paper, we use machine learning algorithms to map requirements to the MITRE ATT&CK database and determine the accuracy of each mapping depending on the data split.

The growing reliance of society on social media for authentic information has done nothing but increase over the past years. This has only raised the potential consequences of the spread of misinformation. One of the growing methods in popularity is to deceive users through the use of a deepfake. A deepfake is a new invention that has come with the latest technological advancements, which enables nefarious online users to replace one's face with a computer-generated, synthetic face of numerous powerful members of society. Deepfake images and videos now provide the means to mimic important political and cultural figures to spread massive amounts of false information. Models that are able to detect these deepfakes to prevent the spread of misinformation are now of tremendous necessity. In this paper, we propose a new deepfake detection schema utilizing two deep learning algorithms: long short-term memory and multilayer perceptron. We evaluate our model using a publicly available dataset named 140k Real and Fake Faces to detect images altered by a deepfake with accuracies achieved as high as 74.7%.

Social media is currently being used by many individuals online as a major source of information. However, not all information shared online is true, even photos and videos can be doctored. Deepfakes have recently risen with the rise of technological advancement and have allowed nefarious online users to replace one’s face with a computer-generated face of anyone they would like, including important political and cultural figures. Deepfakes are now a tool to be able to spread mass misinformation. There is now an immense need to create models that are able to detect deepfakes and keep them from being spread as seemingly real images or videos. In this paper, we propose a new deepfake detection schema using two popular machine learning algorithms; support vector machine and convolutional neural network, along with a publicly available dataset named the 140k Real and Fake Faces to accurately detect deepfakes in images with accuracy rates reaching as high as 88.33%.