Blekinge Institute of Technology
Not a member yet
13576 research outputs found
Sort by
Combating Deepfakes : A Review of Detection Techniques and Their Limitations
No full textDeepfake technology, powered by advanced AI, enables the creation of highly realistic but entirely synthetic audio, video, and image media. While this technology has legitimate applications, it also poses significant security risks. This thesis provides a comprehensive structured literature review and analysis of the risks associated with deepfake technology and critically examines the challenges in currently available detection methods. The study categorizes the main risks into three main categories: Dis- and Misinformation and Manipulation, Identity Theft and Fraud, and Privacy Invasion and Blackmail. It also identifies critical challenges in detection methods including technical limitations and privacy concerns. This thesis contributes to the field by providing a clear understanding of the threat landscape associated with deepfakes, identifying gaps in detection methods, and offering insights for future research
Machine Learning for Accessible Threat Modeling Based on Software Requirements
No full textBackground. Threat modeling is a proven way to prevent costly software vulnerabilities, yet many teams postpone or skip it because manual analysis demands security expertise, time,and money which are in scarce supply. Recent advances in large language models(LLMs) and retrieval-augmented generation (RAG) suggest that portions of this effort could be automated by reasoning directly over natural-language requirements such as user stories. Objectives. This thesis investigates whether an accessible, ML-powered tool can extract action-able security threats from natural language user stories, surface risks that human experts might overlook, and do so with enough accuracy and speed to be useful in an agile workflow. Methods. Following Design Science Research, a two-step artifact was built: a lightweight classifier filters security-irrelevant user stories, after which a RAG-grounded LLM maps the remaining stories to threat database techniques and mitigations. The tool was evaluated on two open-source web projects (51+55 user stories) through quantitative metrics-precision, Exclusive Discovery Rate and qualitative review by a security professional. Results. The best pipeline achieved 83% precision and a 9% Exclusive Discovery Rate, meeting all SMART targets while analyzing 51 user stories in under three minutes on consumer hardware. In several cases the model identified subtle credential-access and discovery vectors initially missed by the expert. Conclusions. Grounded LLMs can reliably transform textual requirements into early-phase threat models, acting as a “second pair of eyes” that lowers the cost and cognitive load of secure-by-design development. While expert validation remains essential, integrating such tools into sprint rituals promises faster, broader, and more consistent threat coverage; future work should generalise beyond web systems and explore interactive,explainable workflows.Bakgrund. Hotmodellering är ett effektivt sätt att motverka dyra sårbarheter i mjukvara, mentas ofta inte på det allvar som behövs. Ofta skjuts det upp, eller ignoreras fullständigt då det krävs tid, pengar och skicklighet för att utföra ett bra arbete. Medhjälp av den stora utvecklingen som skett inom Large Language Models (LLMs) ochRetrieval-Augmented Generation (RAG) så undersöker vi om det är möjligt att utföra effektiv hotmodellering utifrån User Stories med hjälp av LLM-verktyg, och påså sätt göra hotmodellering mer lättillgängligt. Syfte. Denna avhandling undersöker om det är möjligt och lämpligt att med hjälp av ett ML-baserat hjälpmedel utföra hotmodellering. Kan ett ML-baserat verktyg hitta och resonera kring möjliga hot med enbart informationen som finns i ett projekts User Stories, och göra detta med tillräcklig precision för att användas i utveckling? Metod. I enlighet med Design Science Research tillverkades en produkt. Denna produktbestår av två delar: en klassificerare som filtrerar ut User Stories som inte är relevanta för säkerhet, och en LLM som med hjälp av RAG länkar User Stories tillmöjliga attack tekniker från en hotdatabas. Verktyget evaluerades på två projekt med öppen källkod, vilka tillsammans har 106 User Stories. För kvantitativ evaluering användes Precision och Exclusive Discovery Rate (EDR). Kvalitativ evaluering utfördes av en säkerhetsexpert. Resultat. De bästa resultaten som uppnåddes var 83% precision, och 9% EDR. Detta innebäratt verktyget nådde alla SMART-mål uppsatta på en analys av 51 user stories, vilket tog mindre än tre minuter på konsumenthårdvara. Det förekom flera fall då verktyget identifierade hot som säkerhetsexperten till en början hade missat. Slutsatser. LLM:er med RAG kan på ett tillförlitligt sätt hämta information från User Stories och använda denna information för att hitta möjliga hot i ett projekts tidiga skede,och kan användas för att lätta det kognitiva arbetet som krävs av en expert som utför hotmodellering. Validering från mänskiga experter är fortfarande nödvändigt, men verktyg som detta skulle kunna användas för att förbättra kvaliteten på hot-modellering i många utvecklingsprojekt
Ett gestaltningsförslag: Brottsförebyggande åtgärder som främjar trygghetsupplevelser i offentliga miljöer
No full textUnder senare år har ökningen av brottslighet i Sverige lett till en förändrad upplevelse av trygghet i offentliga miljöer, vilket i sin tur har minskat människors benägenhet att vistas i stadsmiljöer, särskilt under kvällstid. Detta tydliggör behovet av trygghetsskapande åtgärder inom stadsplaneringen. Traditionellt har brottsförebyggande arbete i Sverige främst fokuserat på social prevention, insatser som riktar sig mot individens beteende och levnadsförhållanden. Den situationella brottspreventionen, som istället betonar den fysiska miljöns utformning och dess påverkan på brottslighet, har fått mindre utrymme i planeringspraktiken. Denna kunskapsbrist understryker vikten av att stärka kompetensen inom fysisk brottsprevention och att i större utsträckning integrera dessa strategier i planeringen av stadsmiljöer. Genom att systematiskt arbeta med trygghetsskapande principer i den fysiska utformningen kan offentliga miljöer göras mer motståndskraftiga mot brott och samtidigt bidra till en ökad trygghetsupplevelse för invånarna. Syftet med studien är att undersöka hur stadsplanering kan integrera brottsförebyggande åtgärder och bidra till en ökad trygghetskänsla i offentliga miljöer. För att studera detta har en metodkombination tillämpats där valda metoder är dokumentstudie och platsanalys. Platsanalysen baseras på inhämtat underlag från kunskapsöversikten och har gett insikt till det aktuella läget. Dessa metodval har tillämpats i syfte av att studera ämnesområdet ur fler perspektiv samt därmed bidra till bredare kunskap. Studien visar därav hur brottspreventiva åtgärder kan appliceras i fysisk form för att skapa trygghetsfrämjande urbana miljöer.
Att förebygga brott genom den fysiska miljöns utformning : Ett gestaltande projekt för att minska brottsproblematiken och öka tryggheten i en urban miljö
No full textAtt känna sig trygg och vara en del av samhället är en mänsklig rättighet och viktig förutsättning för en hållbar utveckling. Den fysiska planeringen spelar en avgörande roll för den fysiska miljöns utformning. Utformningen av den fysiska miljön kan i sin tur påverka hur stor risken är för att ett brott ska begås vid en plats. Denna uppsats tar utgångspunkt i att känslan av otrygghet grundas i en rädsla för brott. Det betyder att om den fysiska miljön kan utformas på ett sätt som minskar risken för brott, kommer även den upplevda tryggheten att ökas på platsen. Kandidatarbetet syftar till att undersöka hur förekomsten av brott påverkas av den fysiska miljöns utformning. Studien ska ta reda på och förstå vilka faktorer och åtgärder i utformningen av den fysiska miljön som minskar risken för brott. Således kan studien bidra med information om hur dessa åtgärder på ett effektivt sätt kan implementeras i ett verkligt fall för att minska brottsproblematiken och bidra till att färre människor känner sig otrygga i offentliga miljöer. Arbetet avgränsas till att utgå ifrån ett planeringsperspektiv och fokuserar därmed endast på den fysiska miljöns utformning. I uppsatsen genomförs en fallstudie vid Karlskronas stadsbibliotek och dess närområde. Anledningen till att just den här platsen valts är för att kommunens arbete kring brottsbekämpning och deras verktyg EST – Effektiv Samordning för Trygghet pekar ut stadsbiblioteket och dess närområde som den mest otrygga och brottsdrabbade platsen på Trossö, Karlskrona. Informationen är hämtad från en ”heat map” som sammanställer inrapporterade brottsliga och otrygghetsskapande händelser från olika rapportörer som använder verktyget. Studien använder sig av metoderna dokumentär forskning och observationer vid en platsanalys för datainsamling. Observationerna utgår ifrån teorin CPTED - Crime Prevention Through Environmental Design och dess principer som även använts i arbetets designkoncept. CPTED är en teori som påvisat att utformningen av den fysiska miljön och dess koppling till samhället kan användas för att stoppa kriminalitet och öka tryggheten. Platsanalysens resultat visade att flera av de åtgärder som minskar risken för brott var bristfälliga vid den utvalda platsen. Platsanalysen bidrog således med information som förtydligar sambandet mellan den fysiska miljöns utformning och förekomsten av brott. I studien har ett gestaltningsförslag tagits fram som syftar till att visa hur åtgärder i den fysiska miljöns utformning kan implementeras och förbättras på en brottsutsatt plats för att minska risken för brott. Utifrån studiens designkoncept samt tidigare forskning kan det med hög grad av säkerhet förutsägas att gestaltningsförslaget kommer att bidra till att minska risken för brott och andra otrygghetsskapande händelser vid platsen. Kandidatarbetet avslutas med en diskussion gällande empirins resultat samt andra reflektioner som uppstått under arbetets gång
Caching: Investigation into methods for Improving Software Package Updates and Installation success rates
No full textReliable software package installation is critical for maintaining and deploying Linux-based systems. However, failures—particularly under constrained bandwidth or unstable network conditions—pose a significant challenge. This thesis investigates strategies to mitigate such failures through caching, local mirroring, and system-level timeout configurations. We evaluate three distinct network configurations: default APT using external repositories, NFS-mounted repositories, and local HTTP mirrors. Controlled experiments across varying bandwidth levels are conducted to assess installation reliability and performance. Additionally, we modify APT source code to analyze the impact of connection and read timeout settings. Our findings reveal how repository configuration and timeout values influence installation outcomes, providing practical insights to improve package management systems under adverse network conditions
Equity Cost of Capital under Behavioral Distortions : Explaining Jensen’s Alpha in Meme Stocks
No full textTraditional asset‐pricing frameworks like the Capital Asset Pricing Model (CAPM) systematically fail to capture extreme price swings driven by retail‐led “meme” stock frenzies, broader coordination effects, and behavioral biases in financial markets. Jensen’s Alpha is a measure of the abnormal return of an asset relative to its expected CAPM return given its systematic risk (beta), with positive values indicating outperformance and negative values indicating underperformance, making it a direct gauge of CAPM mispricing. This thesis quantifies CAPM’s failure by estimating Jensen’s Alpha through rolling CAPM regressions on daily, weekly, and monthly data from 2014–2023 for five flagship meme stocks and a custom equal‐weighted meme‐stock index (MSI-EW). We complement descriptive statistics of Jensen’s Alpha with correlation analysis and univariate OLS regressions to examine the explanatory power of speculator-driven herding metrics, namely, the average cross-sectional standard deviation (CSSD) and average cross-sectional absolute deviation (CSAD) computed over the same rolling windows as Jensen’s Alpha, both benchmarked to a broad-market ETF (VTI) and to MSI-EW for comparison, alongside trading-volume proxies. Our findings reveal that average CSSD offers stronger explanatory power than average CSAD, with average CSSD benchmarked to the US total-market index ETF (VTI) explaining over 95 % of Alpha variation in GameStop and KOSS across all sampling frequencies, surpassing the explanatory power of average CSSD benchmarked to MSI-EW, while volatility and volume metrics add minimal incremental insight. Time-series visualizations further confirm that spikes in return dispersion coincide with abnormal return episodes. These results illuminate the behavioral drivers behind CAPM pricing errors and introduce a versatile, dispersion-based tool for understanding herding phenomena broadly, empowering businesses and investors to enhance their equity cost of capital calculations and forecasting accuracy in any market environment. Future research could extend this framework by incorporating multivariate behavioral and liquidity factors, leveraging intraday data and sentiment indices, and applying Granger-causality analysis to untangle the directional interplay between dispersion and mispricing
Comparative Analysis of SVM and MobileNetV3Small for Plant Disease Classification : A Study on Classification Accuracy Using SVM and Deep Learning
No full textBackground: Accurate plant disease detection is essential for maintaining agricultural productivity and sustainability. With the rise of machine learning techniques, comparing deep learning models such as MobileNetV3Small with traditional machine learning models like Support Vector Machines (SVM) has become increasingly important in improving plant disease classification accuracy. Objectives: This research aims to compare the performance of SVM and MobileNetV3Small models for the classification of plant diseases, focusing specifically on tomato and watermelon leaf conditions. The goal is to evaluate both models in terms of classification accuracy, interpretability and efficiency. Methods: The study employs two classification models which are SVM, a traditional machine learning algorithm and MobileNetV3Small, a deep learning model. A dataset comprising images of healthy and diseased tomato and watermelon leaves is used. The performance of both models is assessed using metrics such as accuracy, precision, recall and F1 score. Additionally, model interpretability is evaluated through techniques like feature importance analysis and visualization. Results: The MobileNetV3Small model outperforms the SVM model in terms of classification accuracy, achieving higher precision and recall. The deep learning model demonstrates better generalization on test data, while the SVM model offers greater interpretability due to its simpler architecture. Conclusions: While MobileNetV3Small provides superior performance in terms of accuracy, SVM remains a practical option where model interpretability is essential. This comparison highlights the trade-offs between deep learning and traditional machine learning approaches in plant disease classification, offering insights into their respective strengths and applications in agricultural settings
Impact of leadership styles on employee commitment on hybrid working and management modality for remote working
No full textBackground: Hybrid and remote work are not more a trend but a reality that needs to be better understood by the organizations. A deep understanding of how remote work and leadership coexists is needed. Purpose: To investigate which leadership type results in higher employee commitment in hybrid working and which management strategy increases efficiency in remote working Method: Mixed (quantitative and qualitative) Results and analysis: Transactional leadership is perceived as the most effective leadership type by the employees to have higher affective and normative commitment in hybrid working environments. On the other hand, managers tend to adapt laissez-faire leadership to overcome the challenges of remote working conditions. The findings are contradictory with the majority of previous research, which advocates transformational leadership as the most efficient leadership type for hybrid working environments. Other factors (education level, company size, experience) and their interactions are also significant factors which impact employee commitment in hybrid work environments. Conclusions: Employees and managers who participated in this study have different perspectives on which leadership type is the most efficient to have higher employee commitment for hybrid working. Recommendations for future research: Further research is recommended with more participants from different countries and more factors influencing employee commitment can be explored
Identifying Abnormalities in Heart Sound data using Machine Learning
No full textIntroduction: This thesis explores the use of machine learning to automatically detect abnormalities in heart sounds, known as phonocardiograms (PCGs). These sounds, recorded using digital stethoscopes, carry vital information about the mechanical function of the heart. The main objective of this work is to develop an accurate and interpretable system that supports early diagnosis of heart conditions in a practical and clinically relevant way. Related Work: Previous studies have used both traditional machine learning and deep learning for heart sound classification. Traditional models often relied on basic features like zero-crossing rate or spectral centroid, but lacked clinical relevance. Deep learning models improved accuracy but acted as black boxes, offering limited interpretability. Most prior work did not incorporate explainability, which is essential in medical contexts. This thesis addresses these gaps by comparing a feature-based method with a deep learning baseline, focusing on transparency and clinical usefulness. Method: Two approaches were implemented and evaluated. First, a one-dimensional convolutional neural network (ID CNN) was used as a pilot study to assess classification performance and examine the limitations of black-box models. Although Grad-CAM was used to highlight signal regions influencing predictions, the specific clinical relevance such as which features made the model label a heart sound as abnormal remained unclear. To address this, a second, feature-based pipeline was developed. It involved signal denoising, S1/S2 detection, and extraction of clinically relevant features like heart rate, cardiac cycle duration, MFCCs, entropy, and energy. These features were used to train interpretable models including Random Forest, SVM, and XGBoost. LIME was then applied to visualize how each feature influenced the prediction, making it possible to understand what aspect of the heart sound such as frequency patterns in MFCCs or timing irregularities may be contributing to an abnormal classification. All experiments used 3,240 PCG recordings from the PhysioNet database. Results and Analysis: Both approaches successfully classified heart sounds as normal or abnormal. However, the feature-based models offered more transparent reasoning be hind their decisions. With LIME, it became pogible to see which clinically relevant features such as changes in MFCCs or heart rate variability were contributing to a specific prediction, helping us understand what aspect of the heart sound indicated an abnormality. Discussion: This study shows that combining clinical signal features with interpretable machine learning models results in systems that are not only accurate but also understandable and practical for clinical use. It also reveals that while deep learning models like ID CNNs can achieve good accuracy, they lack the transparency needed for medical decision-making unless paired with effective explainability tools. Conclusion: By focusing on explainable classification of PCG recordings, this thesis contributes toward building machine learning systems that support early detection of heart abnormalities in a way that clinicians can trust and act upon with confidence
Unlocking Software Potential Through App Store Reviews
No full textBackground: The software development industry is advancing rapidly, driven by the increasing demand for responsive and innovative applications. Within this landscape, gaming apps stand out due to their complexity and dynamic user expectations, requiring frequent updates and innovative features to maintain user engagement. App store reviews, a largely untapped resource, offer invaluable insights into user experiences and expectations, particularly in the gaming domain. While significant research has focused on descriptive analytics and sentiment analysis, a critical gap exists in systematically extracting and categorizing specific requirement types from unstructured review data for gaming apps. Objectives: The aim of this research is to uncover actionable insights from user reviews of gaming apps to enhance the software development process. The objectives include predicting the sentiment (positive, negative, or neutral) of review comments using NLP models such as TextBlob, BERT, and Large Language Model (LLM); categorizing the comments into distinct categories to facilitate targeted softwareimprovements; and evaluating the performance of these models to determine theiraccuracy and effectiveness in sentiment prediction and feedback categorization. Methods: The implementation methodology for achieving the stated objectives involves a multi-step approach using natural language processing (NLP) and machine learning techniques. First, app store review data is collected and preprocessed to clean and structure the text for analysis. Sentiment prediction models, including TextBlob, BERT, and Large Language Model (LLM), are then applied to classify the reviews as positive, negative, or neutral. For feedback categorization. SemanticClustering is used to classify reviews into distinct categories. Evaluation metrics such as accuracy, precision, recall, and F1-score are used to assess and compare the performance of TextBlob, BERT, and LLM. The insights derived from this analysis are validated through cross-validation and tested for their applicability in guiding software development and enhancement efforts. Results: The performance results of the implemented models demonstrate varying levels of accuracy in sentiment prediction. The TextBlob model, being a lightweight and lexicon-based approach, achieved a performance accuracy of 87%, showcasing its utility for basic sentiment analysis but limited contextual understanding. TheBERT model, leveraging its deep contextual embeddings, improved accuracy to 90%, reflecting its ability to handle complex language nuances and provide more precise classifications. The Large Language Model (LLM) outperformed both, achieving the highest accuracy of 93%, demonstrating their advanced capability in capturing semantic relationships and nuanced text patterns. These results highlight the effectiveness of advanced NLP techniques, with LLM emerging as the most reliable tool for extracting actionable insights from unstructured app store review data. For categorization, the Semantic Clustering method helps in classifying reviews based on the content of reviews with ease by using top keywords and provides valuable insight to developers. Conclusions: This research highlights the potential of NLP and machine learning techniques in deriving actionable insights from unstructured app store review data. By employing TextBlob, BERT, and Large Language Model (LLM), the study successfully predicted review sentiments and categorized feedback into several distinct categories using Semantic Clustering. The results demonstrated that while TextBlob provides a lightweight and moderately accurate solution, BERT and LLM significantly enhance performance, with LLM achieving the highest accuracy of 93%. These findings underscore the effectiveness of advanced NLP models in bridging the gap between user feedback and software development processes, enabling a more user-centered approach to application design and improvement. The methodology and insights from this study can be extended to other domains requiring unstructured text analysis, paving the way for broader applications in data-driven decision-making