http://dspace.epoka.edu.al/handle/1/1666 Sun, 25 Jan 2026 16:55:13 GMT 2026-01-25T16:55:13Z http://dspace.epoka.edu.al/handle/1/2593 A DYNAMIC MODEL FOR PERSONALIZED E-LEARNING USING INTERNET OF THINGS Spaho, Edlir Personalized Online Learning is an adaptive educational approach that tailors learning experiences based on individual learner needs, preferences, and progress. It utilizes artificial intelligence, machine learning, and big data analytics to customize teaching and learning services, thus enhancing academic performance. Traditional POL systems predominantly rely on limited dynamic learner profiles based on behavioral and historical data. They struggle to provide real-time adaptation and context-aware personalization by disregarding critical dynamic factors such as cognitive load, emotional state, and environmental conditions. This dissertation proposes an innovative Internet of Things (IoT) based personalized e-learning framework that integrates real-time learner data, addressing the limitations of current approaches in adaptive e-learning. The research introduces a novel IoT-based dynamic model for personalized e-learning, comprising five modules: (1) a prior knowledge classification and clustering module that dynamically adjusts learning content level based on learners prior knowledge levels; (2) a machine learning-driven learning style preference module that customizes content formats according to individual learning styles; (3) an IoT module that integrates environmental and biological data; (4) a customized Learning Object Container that enriches learning objects with additional metadata; and (5) a rule-based Smart Engine that manages and integrates all modules to personalize learning materials dynamically. The framework is implemented within the Moodle learning management system, and a pilot case study is conducted to evaluate its effectiveness. The study addresses significant challenges, including prior knowledge classification and clustering model, new learning style classification model, real-time data processing, and integration of IoT devices in POL systems. It also explores pedagogical considerations such as cognitive load management, adaptive assessment and feedback, and interactive learning strategies. Empirical validation demonstrates the model’s effectiveness in improving engagement, knowledge retention, and academic performance through data-driven, context-aware adaptation. This research advances the field of IoT-based educational technologies by offering a scalable, intelligent framework that transforms online learning into a dynamic learner-centered experience. Future research directions include artificial intelligence enhanced personalization, scalability in large-scale deployments, and integration with emerging educational technologies. Mon, 28 Apr 2025 00:00:00 GMT http://dspace.epoka.edu.al/handle/1/2593 2025-04-28T00:00:00Z http://dspace.epoka.edu.al/handle/1/2592 BIOMATERIAL TOXICITY ASSESSMENT USING NON-INVASIVE MICROSCOPY AND AI-DRIVEN CELL MORPHOLOGY ANALYSIS Duro, Xhoena Medical image analysis has significantly advanced with machine learning, enhancing disease diagnosis and biomaterial toxicity assessment. However, challenges such as data heterogeneity, computational complexity, and the lack of standardized evaluation metrics hinder its full potential. The study investigates supervised classification of three morphologically similar cell types, evaluating architectures (VGG16, Inception V3, SqueezeNet) and classifiers (Neural Network, Random Forest, KNN, etc.). Results indicate VGG16 paired with Neural Networks achieves the highest accuracy. Unsupervised clustering is explored by applying ISO guidelines to assess biomaterial toxicity levels, leveraging VGG16 and SqueezeNet for feature extraction. A hybrid clustering approach enhances classification into toxicity levels, demonstrating improved separability with high-pass filtering techniques. A U-Net-based model is optimized for cell counting, evaluating optimizer-loss function combinations for segmentation and confluency analysis. v Experiments on cells exposed to biomaterials (PAR50, UniFast) reveal toxicity patterns through morphological changes. Hybrid loss functions (Dice-Focal, Jaccard-BCE) significantly improve segmentation accuracy. Quantization and pruning techniques are applied to reduce computational demands without compromising accuracy to enable real-world deployment. A pruned U-Net achieves 95% segmentation accuracy. This research contributes novel methodologies for biomedical image analysis by: (i) developing a benchmarked unsupervised clustering framework aligned with ISO standards, (ii) proposing a high-accuracy classification model for cell types, (iii) optimizing U-Net for segmentation and counting, and (iv) enhancing computational efficiency through model compression. These findings support automated biomaterial toxicity assessment, improving efficiency and standardization in medical imaging applications. Tue, 15 Apr 2025 00:00:00 GMT http://dspace.epoka.edu.al/handle/1/2592 2025-04-15T00:00:00Z http://dspace.epoka.edu.al/handle/1/2120 ANALYZING CUSTOMER REVIEWS IN TURKISH USING MACHINE LEARNING AND DATA SCIENCE METHODOLOGIES Ceyhan, Migena Digital life, especially after the introduction of Web 2.0, has significantly altered human relations, providing all people the “right of public speech”. Ideas, emotions, and opinions on many topics are generously shared in virtual environments. A new age global and digital Mouth of World is shaping the society where knowledge is the most influential power. Being fed by social media data highly dynamic in either amount or shape, automatic handling is indispensable. Natural Language Processing, in cooperation with Machine Language techniques, has an important say in analyzing written textual data. Traditional techniques exploited in the literature are empowered when hybrid ones are applied, in accordance also with the characteristic properties of the language used and the domain-specific data. Although all the subsequent steps of the text classification chain are important, adequate feature selecting has a notable huge impact on accurate classification prediction. In this study, a simple classification of the sentiment polarity of comments in document level of subjective texts in Turkish is done. Different domains include reviews of customers towards company products, movies, and healthcare services, deciding on the positivity or negativity of the comments. Another domain includes doctors’ notes on patients’ symptoms aiming to predict and thus recommend some of the most often used medical tests according to general doctors’ procedures. The features used included a part of or all distinct words roots together with their binary or frequency information. Linear or vector analysis of the feature sets was done employing Machine Learning algorithms provided by the Weka tool. Hybrid features set was proposed and found more efficient combining binary vectors and frequency meta-features from nodes and leaves of J48 tree classifier for all or a set of correlation based selected features, improving both prediction accuracy and classification performance. Mon, 25 Jan 2021 00:00:00 GMT http://dspace.epoka.edu.al/handle/1/2120 2021-01-25T00:00:00Z http://dspace.epoka.edu.al/handle/1/1887 A ROAD TRAFFIC VIOLATION DETECTION & REPORTING SYSTEM ÖZKUL, MÜKREMIN This study presents a police-less multi-party traffic violation detection and reporting system, that does not rely on costly infrastructure or the presence of law enforcement. It relies solely on broadcast messages among vehicles and report delivery to the transportation authority. Firstly, a vehicle is modeled as an automaton (in computational sense) that has its own state and has a read access to the state of other automata of other vehicles in a neighborhood of fixed size. The common traffic rules and communication rules make the program of these automata that guide the transitions of the vehicles in space and time. By observing the transitions of the vehicles in their neighborhood, a vehicle can decide if these comply with the traffic rules encoded in the system. Whenever a transition is not performed according to the program, a violation occurs. These violations are reported and witnessed to the transportation authority by the vehicles in the neighborhood which act as witnesses and reporters. The system is able find the location and real identity of any vehicle whenever it commits a rule violation in traffic with a lightweight protocol. Yet, the system preserves privacy and allows no false positives. Tue, 17 Dec 2019 00:00:00 GMT http://dspace.epoka.edu.al/handle/1/1887 2019-12-17T00:00:00Z