Search

Large language models (LLMs) have transformed Natural Language Processing (NLP) by enabling robust text generation and understanding. However, their deployment in sensitive domains like healthcare, finance, and legal services raises critical concerns about privacy and data security. This paper proposes a comprehensive framework for embedding trust mechanisms into LLMs to dynamically control the disclosure of sensitive information. The framework integrates three core components: User Trust Profiling, Information Sensitivity Detection, and Adaptive Output Control. By leveraging techniques such as Role-Based Access Control (RBAC), Attribute-Based Access Control (ABAC), Named Entity Recognition (NER), contextual analysis, and privacy-preserving methods like differential privacy, the system ensures that sensitive information is disclosed appropriately based on the user's trust level. By focusing on balancing data utility and privacy, the proposed solution offers a novel approach to securely deploying LLMs in high-risk environments. Future work will focus on testing this framework across various domains to evaluate its effectiveness in managing sensitive data while maintaining system efficiency. [ABSTRACT FROM AUTHOR]

This study investigates the effectiveness of Apple's privacy labels, introduced in iOS 14, in promoting transparency around app data collection practices with respect to the GDPR. Specifically, we address two key research questions: (1) What special categories of personal data, as regulated by the GDPR, are collected and used by apps, and for which purposes? (2) What disparities exist between app-stated permissions and the apparent unnecessary data gathering across various categories in the iOS App Store? By analyzing a comprehensive dataset of 541,662 iOS apps, we identify common practices related to prevalent use of sensitive and special categories of personal data, revealing widespread instances of unnecessary data collection, misuse, and potential GDPR violations. Furthermore, our analysis uncovers significant inconsistencies between the permissions stated by apps and the actual data they gather, highlighting a critical gap in user privacy protection within the iOS ecosystem. These findings underscore the need for stricter regulatory oversight of app stores and the necessity of effective privacy notices to build accountability and trust and ensure transparency. This study offers actionable insights for regulators, app developers, and users towards creating secure and transparent digital ecosystems. [ABSTRACT FROM AUTHOR]

This exploratory study is based on semi-structured interviews with digital preservationists working in Canada. The goal was to understand how participants are reviewing sensitive personal information in their digital materials and identify the challenges they face in this work. Findings include a summary of current practices in place (e.g., risk profiling, using software tools for triage, and consultation with donors and/or community), and a list of challenges (e.g., IT or systems infrastructure, scale, barriers to access for software tools, and funding restrictions). The implications of these findings are that Canadian memory workers require time, staff, support, and training, while developers of assistive-software tools for this work should address tool limitations and prioritize user-friendly interfaces. Grant funding stipulations should also be revised to allow for diverse access options, while investment in controlled access repositories would enable responsible online access to complex collections. [ABSTRACT FROM AUTHOR]

Social network services and chatbots are susceptible to personal information leakage while facilitating language learning without time or space constraints. Accurate detection of personal information is paramount in avoiding such leaks. Conventionally named entity recognizers commonly used for this purpose often fail owing to errors of unrecognition and misrecognition. Research in named entity recognition predominantly focuses on English, which poses challenges for non-English languages. By specifying procedures for the development of Korean-based tag sets, data collection, and preprocessing, we formulated directions on the application of entity recognition research to non-English languages. Such research could significantly benefit artificial intelligence (AI)-based natural language processing globally. We developed a personal information tag set comprising 33 items and established guidelines for dataset creation, later converting it into JSON format for AI learning. State-of-the-art AI models, BERT and ELECTRA, were employed to implement and evaluate the named entity recognition (NER) model, which achieved an 0.943 F1-score and outperformed conventional recognizers in detecting personal information. This advancement suggests that the proposed NER model can effectively prevent personal information leakage in systems processing interactive text data, marking a significant stride in safeguarding privacy across digital platforms. [ABSTRACT FROM AUTHOR]

A novel use of visual cryptography is to ensure safe bio-metric authentication in the healthcare industry. The necessity for strong authentication techniques is due to the growing digitization of records and the growing reliance on electronic health systems. There are now a lot of obstacles because of security and privacy issues with conventional bio-metric authentication techniques.Visual cryptography can be done using many algorithms for image encryption. This paper presents a method to encrypt images using two main encryption algorithms: the Advanced Encryption Standard and the Shamir Secret Sharing Scheme. AES, a widely used symmetric encryption algorithm, provides high security by converting plain text to ciphered text using a key which is symmetric. By using the Shamir algorithm, the system creates and saves encrypted images using visual cryptography. In the health care industries, biometric samples are presented as part of the authentication procedure; these samples are divided into shares and encrypted. The shares are subsequently given to the healthcare professionals, who authenticate patients using their unique shares and the help of the central server. By preventing the disclosure of any critical information to a single party, the visual cryptography technology reduces the possibility of data breaches. [ABSTRACT FROM AUTHOR]

Data Breach is a security problem in which bad guys gain access to confidential or sensitive information along with personal data like social security numbers, bank account numbers, healthcare data or corporate data which include customer data records and financial information. It can be accidental or purposely exposed to people. Nowadays, data is super important for businesses. If data leaks out, it can really be a big delinquent to a company's reputation and a big financial loss. With more and more data being created, and data breaches happening a lot, stopping data leaks is a big challenge. This review helps readers to learn about data leaks, threats, recent data leaks and various prevention techniques. [ABSTRACT FROM AUTHOR]

Increasing numbers of social media users report being blackmailed on social media following their sharing of sensitive and personal information. However, little is known about this emerging crime and its victims. Using a survey of adult WhatsApp users in Oman (n = 1452) and interviews (n = 18) with victims and Omani criminal justice professionals, this study addresses this gap by examining the prevalence of blackmail among adult WhatsApp users arising from their voluntary self-disclosures, as well as the characteristics and vulnerabilities of those who fall victim to this activity. New insights are also offered into the factors influencing victim's decisions to report their experiences to criminal justice professionals, as well as the challenges associated with investigating and prosecuting this activity. [ABSTRACT FROM AUTHOR]

This study is based on semi-structured interviews with digital preservation practitioners working in Canada. Participants included librarians, archivists, a library director, and technical systems staff. The goal was to understand how participants are reviewing for sensitive personal information in their digital materials, what they are doing with that information, and identify the challenges they face in this work. Qualitative research methods used leaned heavily on feminist methodologies (Stanley & Wise; Taylor) and the field of ethnography (Feldman, Bell & Berger; Emerson, Fretz & Shaw). Findings include a summary of current strategies in place (e.g., risk profiling, using software tools for triage, and consultation with donors and/or community), a list of challenges (e.g., IT or systems infrastructure, barriers to access for software tools, funding restrictions, and scale), and a series of recommendations (e.g., increasing staff resources, developing tools geared towards triage, and improving grant requirements to recognize this work).

Privacy concerns have led customer-centric companies to adopt fair information practices to protect their customers' private information. This situation forces organizations to understand how their consumers perceive their private data when they are disclosed or are already known by the company. This study aims to establish a classification of private data based on customer perceptions to allow companies to address rising privacy concerns. This study developed a two-dimensional map of the positioning of various pieces of information according to customer perceptions. A total of 157 observations were collected online, and multidimensional scaling was used as an analysis technique. The dimensions that explain the resulting classification or spatial map were the degree of exposure customers feel when their data are known and the likelihood that a specific piece of information can identify a particular person. [ABSTRACT FROM AUTHOR]

Objectives. As the scope of personal data transmitted online continues to grow, national legislatures are increasingly regulating the storage and processing of digital information. This paper raises the problem of protecting personal data and other confidential information such as bank secrecy or medical confidentiality of individuals. One approach to the protection of confidential data is to depersonalize it, i.e., to transform it so that it becomes impossible to identify the specific subject to whom the data belongs. The aim of the work is to develop a method for the rapid and safe automation of the depersonalization process using machine learning technologies.Methods. The authors propose the use of artificial intelligence models to implement a system for the automatic depersonalization of personal data without the use of human labor to preclude the possibility of recognizing confidential information even in unstructured data with sufficient accuracy. Rule-based algorithms for improving the precision of the depersonalization system are described.Results. In order to solve this problem, a model of named entity recognition is trained on confidential data provided by the authors. In conjunction with rule-based algorithms, an F1 score greater than 0.9 is achieved. For solving specific depersonalization problems, a choice between several implemented anonymization algorithm variants can be made.Conclusions. The developed system solves the problem of automatic anonymization of confidential data. This opens an opportunity to ensure the secure processing and transmission of confidential information in many areas, such as banking, government administration, and advertising campaigns. The automation of the depersonalization process makes it possible to transfer confidential information in cases where it is necessary, but not currently possible due to legal restrictions. The distinctive feature of the developed solution is that both structured data and unstructured data are depersonalized, including the preservation of context.

Social network attackers leverage images and text to disseminate sensitive information associated with pornography, politics, and terrorism,causing adverse effects on society.The current sensitive information classification model does not focus on feature fusion between images and text, greatly reducing recognition accuracy.To address this problem, we propose an attentive cross-modal fusion model (ACMF), which utilizes mixed attention mechanism and the Contrastive Language-Image Pre-training model.Specifically, we employ a deep neural network with a mixed attention mechanism as a visual feature extractor. This allows us to progressively extract features at different levels. We combine these visual features with those obtained from a text feature extractor and incorporate image-text frequency domain information at various levels to enable fine-grained modeling. Additionally, we introduce a cyclic attention mechanism and integrate the Contrastive Language-Image Pre-training model to establish stronger connections between modalities, thereby enhancing classification performance.Experimental evaluations conducted on sensitive information datasets collected demonstrate the superiority of our method over other baseline models. The model achieves an accuracy rate of 91.4% and an F1-score of 0.9145. These results validate the effectiveness of the mixed attention mechanism in enhancing the utilization of important features. Furthermore, the effective fusion of text and image features significantly improves the classification ability of the deep neural network. [ABSTRACT FROM AUTHOR]

Privacy preserving utility mining (PPUM) aims to solve the problem of sensitive information leakage in utility pattern mining. In recent years, researchers have proposed algorithms to solve the privacy-preserving problem. However, these algorithms have high side effects, long sanitization time, and computational complexity. Although the FPUTT algorithm reduces the number of database scans, tree construction and traversal still take much time. The paper proposes a fast utility-list dictionary algorithm (FULD). The utility-list dictionary consists of all sensitive items. Through dictionary lookup, sensitive items can be found and modified. In addition, the novel concepts of SINS and tns are proposed to reduce the side effects of the algorithm. In this paper, the experiments show that the FULD algorithm has good performance, such as running time and side effects. The running time of the FULD is 15–20 times shorter than the FPUTT algorithm. It performs well both on sparse and dense datasets. [ABSTRACT FROM AUTHOR]

Social network attackers leverage images and text to disseminate sensitive information associated with pornography, politics, and terrorism,causing adverse effects on society.The current sensitive information classification model does not focus on feature fusion between images and text, greatly reducing recognition accuracy.To address this problem, we propose an attentive cross-modal fusion model (ACMF), which utilizes mixed attention mechanism and the Contrastive Language-Image Pre-training model.Specifically, we employ a deep neural network with a mixed attention mechanism as a visual feature extractor. This allows us to progressively extract features at different levels. We combine these visual features with those obtained from a text feature extractor and incorporate image-text frequency domain information at various levels to enable fine-grained modeling. Additionally, we introduce a cyclic attention mechanism and integrate the Contrastive Language-Image Pre-training model to establish stronger connections between modalities, thereby enhancing classification performance.Experimental evaluations conducted on sensitive information datasets collected demonstrate the superiority of our method over other baseline models. The model achieves an accuracy rate of 91.4% and an F1-score of 0.9145. These results validate the effectiveness of the mixed attention mechanism in enhancing the utilization of important features. Furthermore, the effective fusion of text and image features significantly improves the classification ability of the deep neural network. [ABSTRACT FROM AUTHOR]

The Unified Extensible Firmware Interface (UEFI) provides a specification of the software interface between an OS and its underlying platform firmware. UEFI UI is an interactive interface that allows users to configure and manage UEFI settings, which is closely related to HII (Human Interface Infrastructure). In practice, HII provides a mechanism that allows developers to create UI elements with HII-related protocols. In this paper, we provide a comprehensive analysis of the UEFI combined with a case study. We proposed a protocol-centered static analysis method to obtain UEFI's password policy, using HII-related protocols to find password implementation. Existing static analyses are ineffective in detecting such password policy in stripped UEFI firmware images. By reverse-engineering the IFR (Internal Forms Representation) in HII, we located where much sensitive information is stored. Lastly, we studied hardware port configurations, using Secure Boot as a case in point. We analyzed how UEFI uses the HII protocol to set relevant information in the UEFI UI. This paper is the first to offer a reverse-engineering systematic analysis of exploring UEFI via HII, providing valuable insights into its structure and potential enhancements for firmware security. [ABSTRACT FROM AUTHOR]

Social network services and chatbots are susceptible to personal information leakage while facilitating language learning without time or space constraints. Accurate detection of personal information is paramount in avoiding such leaks. Conventionally named entity recognizers commonly used for this purpose often fail owing to errors of unrecognition and misrecognition. Research in named entity recognition predominantly focuses on English, which poses challenges for non-English languages. By specifying procedures for the development of Korean-based tag sets, data collection, and preprocessing, we formulated directions on the application of entity recognition research to non-English languages. Such research could significantly benefit artificial intelligence (AI)-based natural language processing globally. We developed a personal information tag set comprising 33 items and established guidelines for dataset creation, later converting it into JSON format for AI learning. State-of-the-art AI models, BERT and ELECTRA, were employed to implement and evaluate the named entity recognition (NER) model, which achieved an 0.943 F1-score and outperformed conventional recognizers in detecting personal information. This advancement suggests that the proposed NER model can effectively prevent personal information leakage in systems processing interactive text data, marking a significant stride in safeguarding privacy across digital platforms.

With the ever-increasing number of devices, the Internet of Things facilitates the connection between the devices in the hyper-connected world. As the number of interconnected devices increases, sensitive data disclosure becomes an important issue that needs to be addressed. In order to prevent the disclosure of sensitive data, effective and feasible privacy preservation strategies are necessary. A noise-based privacy-preserving model has been proposed in this article. The components of the noise-based privacy-preserving model include Multilevel Noise Treatment for data collection; user preferences-based data classifier to classify sensitive and non-sensitive data; Noise Removal and Fuzzification Mechanism for data access and user-customized privacy preservation mechanism. Experiments have been conducted to evaluate the performance and feasibility of the proposed model. The results have been compared with existing approaches. The experimental results show an improvement in the proposed noise-based privacy-preserving model in terms of computational overhead. The comparative analysis indicates that the proposed model without the fuzzifier has around 52–77% less computational overhead than the Data access control scheme and 46–70% less computational overhead compared to the Dynamic Privacy Protection model. The proposed model with the fuzzifier has around 48–73% less computational overhead compared to the Data access control scheme and 31–63% less computational overhead compared to the Dynamic Privacy Protection model. Furthermore, the privacy analysis has been done with the relevant approaches. The results indicate that the proposed model can customize privacy as per the users' preferences and at the same time takes less execution time which reduces the overhead on the resource constraint IoT devices. [ABSTRACT FROM AUTHOR]

The rapid growth of cloud storage brings forth significant difficulties in ensuring the integrity of data owners' sensitive data. Multi-cloud storage systems, which motive to enhance trustworthiness, are vulnerable to rigorous replacement attacks. These insidious attacks involve malicious actors substituting legitimate data with counterfeit versions, leading to compromised integrity and fraudulent audit outcomes. We propose a novel identity-based data integrity auditing protocol tailored for multi-cloud storage environments to tackle this challenge. Our protocol leverages a Third-Party Auditor (TPA) to delegate data authenticity and auditing responsibilities, employing identity-based blind signatures with a robust data recovery scheme. By eliminating the need for cumbersome public key certificate management, our solution effectively protects data owners' sensitive data while ensuring confidentiality across multiple cloud providers, thereby mitigating the risk of replacement attacks. A thorough security examination shows the proposed system resists forging, replacement attacks, and data confidentiality violations. A rigorous performance experiment also reveals outstanding computational and communication efficiency that outperforms current methods in tag production, proof generation, and verification. Our protocol offers an extraordinary balance of security and efficiency, making it particularly useful for auditing patent-sensitive data in multi-cloud storage. Our proposal provides a timely and effective way to protect sensitive data and prevent attackers in multi-cloud storage settings at a time when replacement attacks pose a serious concern.

In traditional centralized Android malware classifiers based on machine learning, the training sample uploaded by users contains sensitive personal information, such as app usage and device security status, which will undermine personal privacy if used directly by the server. Federated-learning-based Android malware classifiers have attracted much attention due to their privacy-preserving and multi-party joint modeling. However, research shows that indirect privacy inferences from curious central servers threaten this framework. We propose a privacy risk evaluation framework, FedDroidMeter, based on normalized mutual information in response to user privacy requirements to measure the privacy risk in FL-based malware classifiers. It captures the essential cause of the disclosure of sensitive information in classifiers, independent of the attack model and capability. We performed numerical assessments using the Androzoo dataset, the baseline FL-based classifiers, the privacy-inferred attack model, and the baseline methodology of privacy evaluation. The experimental results show that FedDroidMeter can measure the privacy risks of the classifiers more effectively. Meanwhile, by comparing different models, FL, and privacy parameter settings, we proved that FedDroidMeter could compare the privacy risk between different use cases equally. Finally, we preliminarily study the law of privacy risk in classifiers. The experimental results emphasize the importance of providing a systematic privacy risk evaluation framework for FL-based malware classifiers and provide experience and a theoretical basis for studying targeted defense methods. [ABSTRACT FROM AUTHOR]

Pakistan is among the few countries generating and receiving enormous e-waste, which posits a threat to its future generations. A systematic literature review also suggests exploring e-waste awareness in Asia to understand awareness and behavior. Therefore, the present study explored university students' awareness of e-waste and the factors hindering the disposal of laptops, personal computers, and cellphones and suggested a conceptual framework. The study used the qualitative research approach and non-probability sampling. We collected data through four focus group discussions (FGDs) with students at a Pakistani university. After data saturation, we developed themes from FGDs and found computer sciences and engineering students with better awareness than others. The factors hindering e-waste disposal were lower monetary benefits for disposal, breach of sensitive information, nostalgic association with devices, and non-availability of disposal facilities. Other variables like lower resale value and high family sharing increased the storage of e-devices and reduced e-waste disposal. The research is among a few initial attempts to explore e-waste awareness and factors hindering disposal behavior in e-waste-receiving countries (e.g., Pakistan) and provides evidence from students who are the primary users. Our findings are crucial for policymakers to take corrective actions, introduce monetary benefits, and secure disposal to reduce e-waste. Research highlights: We explored the university students' e-waste awareness level and factors inducing the sale/storage of e-devices (management perspective). Student e-waste awareness level is low; only computer science students are aware. Monetary benefits to recycling and family sharing can reduce e-waste generation. [ABSTRACT FROM AUTHOR]

In the big data era, a tremendous volume of electronic documents is transmitted via the network, many of which include sensitive information about the country and businesses. There is a pressing need to be able to perform intelligent sensing of sensitive information on these documents in order to be able to discover and guarantee the security of sensitive information in this enormous volume of documents. Although the low effectiveness of manual detection is resolved by the current method of handling sensitive information, there are still downsides, such as poor processing effects and slow speed. This study creatively proposes the Text Sensitive Information Intelligent Perception algorithm (TSIIP), which detects sensitive words at the word level and sensitive statements at the statement level to obtain the final assessment score of the text. We experimentally compare this algorithm with other methods on an existing dataset of sensitive Chinese information. We use the metrics measuring the accuracy of the binary classification model, where the F1 score reaches 0.938 (+0.6%), and the F2 score reaches 0.946 (+1%), and the experimental results fully demonstrate the superiority of this algorithm. [ABSTRACT FROM AUTHOR]

With the progress of technology in modern hospitals, an intelligent perioperative situation recognition will gain more relevance due to its potential to substantially improve surgical workflows by providing situation knowledge in real-time. Such knowledge can be extracted from image data by machine learning techniques but poses a privacy threat to the staff’s and patients’ personal data. De-identification is a possible solution for removing visual sensitive information. In this work, we developed a YOLO v3 based prototype to detect sensitive areas in the image in real-time. These are then deidentified using common image obfuscation techniques. Our approach shows that it is principle suitable for de-identifying sensitive data in OR images and contributes to a privacyrespectful way of processing in the context of situation recognition in the OR.

Data integrity auditing provides a method for checking the integrity of outsourced data in cloud storage. However, outsourced data often contain sensitive information (such as names), posing risks of exposure during data sharing. To address this issue, Ming et al. proposed a certificateless integrity auditing scheme for sensitive information protection, claiming its security. However, by demonstrating two specific attack scenarios, we pointed out its security vulnerabilities. Subsequently, we proposed a new certificateless integrity auditing scheme for sensitive information protection in cloud storage (CIAS-SIP), which supports sensitive information protection and does not specify the data blocks that need sanitization by the data owner (DO). In addition, it supports dynamic operations by the DO on outsourced data (insertion, deletion, and modification) and provides security proofs based on the discrete logarithm problem. Finally, we compared CIAS-SIP's performance with three other integrity auditing schemes for sensitive information protection. The results show that CIAS-SIP exhibits superior efficiency. [ABSTRACT FROM AUTHOR]

The term cybersecurity refers to an environment capable of protecting digital devices, networks and information from unauthorized access and preventing data theft or alteration. It is composed of a collection of carefully crafted techniques, processes, and practices to protect sensitive information and deterring cyber-attacks. In the recent period, the domain of cybersecurity has undergone rapid growth in response to the increasing cyber threats. Cybersecurity includes important tactics that help protect the digital environment, which are firewalls, encryption, secure passwords, and threat detection and response systems. Employees must be trained on these tactics. This article will discuss the five most pressing challenges facing the cybersecurity industry today that must be taken into account by businesses, organizations, and individuals in order to secure their confidential data from cybercrime. The conclusion of the article highlighted the significance of growing awareness about cybersecurity risks in order to effectively handle digital environments and protect them from any electronic threats. [ABSTRACT FROM AUTHOR]

Copyright of Polish Statistician / Wiadomości Statystyczne is the property of State Treasury - Statistics Poland and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

A game measurement model considering the attacker's knowledge background is proposed based on the Bayesian game theory aiming at striking a balance between the protection of sensitive information and the quality of service. We quantified the sensitive level of information according to the user's personalized sensitive information protection needs. Based on the probability distribution of sensitive level and attacker's knowledge background type, the strategy combination of service provider and attacker was analyzed, and a game-based sensitive information protectionmodelwas constructed. Through the combination of strategies under Bayesian equilibrium, the information entropy was used to measure the leakage of sensitive information. Furthermore, in the paper the influence of the sensitive level of information and the attacker's knowledge background on the strategy of both sides of the game was considered comprehensively. Further on, the leakage of the user's sensitive information was measured. Finally, the feasibility of the model was described by experiments. [ABSTRACT FROM AUTHOR]

Qualitative researchers often engage in work addressing challenging, difficult, or sensitive topics and are consequently exposed to the participants' narratives which may be emotionally charged, distressing, or compromising. These narratives occasionally rest heavy on a researcher's conscience or may linger in the mind. Much literature has assessed how best to keep participants safe, but less attention has been given to how we keep researchers safe. We therefore document the following: (1) Our experiences of the issues presented by undertaking qualitative research involving challenging, difficult, or sensitive topics; and (2) Practical principles devised to overcome these issues, ensuring safety and wellbeing amongst researchers engaging in these types of qualitative research. We provide guidance for qualitative researchers of all levels of experience and expertise on how best to protect and support themselves, their colleagues, and other collaborating research staff, when undertaking qualitative research which might otherwise feel uncomfortable or overwhelming to tackle. [ABSTRACT FROM AUTHOR]

Focus groups (FGs) and individual interviews (IDIs) can be conducted in-person or in several different online contexts. We conducted a quasi-experimental study and assessed sharing of sensitive or dissenting information and participant comfort in FGs and IDIs across four modalities: (1) in-person, (2) online video-based, (3) online chat-based (synchronous), and (4) online email/message board-based (asynchronous). Participants were systematically assigned to one of the four modalities and randomized to one of 24 FGs or 48 IDIs (N = 171). The study topic was medical risk during pregnancy. All participants also completed a survey on their perceptions of the data collection process. We found no significant difference in the frequency of disclosure of sensitive information by modality. Text-based FGs (chat and message board) were more likely to contain dissenting opinions than visually-based FGs (in-person and video). Participants also reported feeling less rapport and personal comfort in sharing information in the FG video modality than other modalities. These findings provide initial data that can guide researchers in choosing among data collection modalities to maximize participant engagement and comfort. [ABSTRACT FROM AUTHOR]

Abstract With the ever-increasing number of devices, the Internet of Things facilitates the connection between the devices in the hyper-connected world. As the number of interconnected devices increases, sensitive data disclosure becomes an important issue that needs to be addressed. In order to prevent the disclosure of sensitive data, effective and feasible privacy preservation strategies are necessary. A noise-based privacy-preserving model has been proposed in this article. The components of the noise-based privacy-preserving model include Multilevel Noise Treatment for data collection; user preferences-based data classifier to classify sensitive and non-sensitive data; Noise Removal and Fuzzification Mechanism for data access and user-customized privacy preservation mechanism. Experiments have been conducted to evaluate the performance and feasibility of the proposed model. The results have been compared with existing approaches. The experimental results show an improvement in the proposed noise-based privacy-preserving model in terms of computational overhead. The comparative analysis indicates that the proposed model without the fuzzifier has around 52–77% less computational overhead than the Data access control scheme and 46–70% less computational overhead compared to the Dynamic Privacy Protection model. The proposed model with the fuzzifier has around 48–73% less computational overhead compared to the Data access control scheme and 31–63% less computational overhead compared to the Dynamic Privacy Protection model. Furthermore, the privacy analysis has been done with the relevant approaches. The results indicate that the proposed model can customize privacy as per the users’ preferences and at the same time takes less execution time which reduces the overhead on the resource constraint IoT devices.

In traditional centralized Android malware classifiers based on machine learning, the training sample uploaded by users contains sensitive personal information, such as app usage and device security status, which will undermine personal privacy if used directly by the server. Federated-learning-based Android malware classifiers have attracted much attention due to their privacy-preserving and multi-party joint modeling. However, research shows that indirect privacy inferences from curious central servers threaten this framework. We propose a privacy risk evaluation framework, FedDroidMeter, based on normalized mutual information in response to user privacy requirements to measure the privacy risk in FL-based malware classifiers. It captures the essential cause of the disclosure of sensitive information in classifiers, independent of the attack model and capability. We performed numerical assessments using the Androzoo dataset, the baseline FL-based classifiers, the privacy-inferred attack model, and the baseline methodology of privacy evaluation. The experimental results show that FedDroidMeter can measure the privacy risks of the classifiers more effectively. Meanwhile, by comparing different models, FL, and privacy parameter settings, we proved that FedDroidMeter could compare the privacy risk between different use cases equally. Finally, we preliminarily study the law of privacy risk in classifiers. The experimental results emphasize the importance of providing a systematic privacy risk evaluation framework for FL-based malware classifiers and provide experience and a theoretical basis for studying targeted defense methods.