Nowadays, data is continuously generated by thousands of data sources which unceasingly send sequences of data records as data streams that include a wide variety of elements of numerical as well as categorical type. Accordingly, these mixed data streams are considered as the main source of big data. With the progress of artificial intelligence and machine learning technologies, systems with acceleration of the decision-making process are required in today’s companies. This makes it challenging in different machine learning algorithms in terms of analyzing massive mixed datasets faster and more accurately. In this context, our aim is to manage a dynamic scope through which new instances escorted with newly added mixed features are forthcoming. To do so, incremental and decremental learning is the key of this challenge. Particularly, we devote this thesis to deeply study the unsupervised k-prototypes clustering algorithm which is able to handle mixed unlabeled data. As data streams proceed, we aim to develop a dynamic k-prototypes clustering method for handling the incremental and decremental attribute, object, and class learning spaces at the same deal. Moreover, not all attributes present in a dataset are relevant for mining and decision-making. Therefore, feature selection preprocessing technique is required for better analysis and deep data interpretation which consists in selecting a subset of only relevant attributes for mining among all original ones. Consequently, this Thesis will also focus on applying our dynamic real-time method in medicine. It would be interesting and prominent not only to understand recent diseases and therapies, but also to predict outcomes at earlier stages and to make real-time decisions and to make people’s lives advantageous and healthier. This proposal presents encouraging experimental results compared to the batch k-prototypes method and number of similar methods. The obtained clusters’ inertia and run time results emphasize the scalability and the performance of our proposed dynamic k-prototypes for mixed data stream clustering.

Recent advances in large language models (LLMs) have enabled new possibilities for automated radiology reporting, yet key challenges remain, including lack of contextualization, absence of longitudinal reasoning, and risk of clinically inaccurate content. We propose a lightweight, modular architecture that combines a pre-trained LLM with a context-aware Prompt Constructor and a temporal reasoning engine. Prompts are dynamically adapted based on imaging modality, clinical indication, and prior reports, supported by a fuzzy logic-driven rule base and a temporal summarizer to ensure clinical nuance and continuity. Our system is evaluated through realistic case scenarios across multiple imaging types and externally validated on 250 cases from the MIMIC-CXR dataset. Results demonstrate significant gains in BLEU, ROUGE, and BERTScore over baseline prompting, with expert review confirming improved structure, diagnostic alignment, and reduced hallucinations. These findings support the practical integration of explainable and configurable LLMbased systems into real-world radiology workflows.

This study investigates the integration of Process Mining (PM) techniques with Machine Learning (ML) algorithms to detect corrupt activities in business processes. PM has gained significant attention for its ability to analyze event logs and uncover inefficiencies, deviations, non-compliance, and regulatory breaches in real processes. However, its application in detecting corruption, fraud, or other unethical practices in organizational processes remains underexplored. Through a structured analysis of existing research, we examine how PM and ML have been applied in related areas such as fraud detection, and evaluate their relevance to addressing corruption-specific challenges. This paper advocates for the use of PM methods combined with ML techniques to improve corruption detection systems, outlining the key challenges and gaps in current approaches. By synthesizing insights from the literature and evaluating use-case applicability, this work provides a foundation for future research into corruption-aware process analytics.

Healthy People 2030 focuses on improving health care quality and making sure that all people get the healthcare services they need. Current healthcare facilities used Electronic Health Record (EHR) systems to manage and store medical records of the patients. There is also no standard inter-connection to store and retrieve medical records of patients among the majority of the hospitals. One of the technologies used to improve the quality of health care is Internet of Things (IoT). It has provided us with the capabilities to improve the standards of healthcare. In this paper, we propose a novel system called Electronic Health Records based on the Internet of Things (EHRIoT) to centralize patient data and improve the standards and the quality of care. By leveraging IoT technology, EHRIoT connects medical resources and enables effective collaboration, providing smart healthcare services to patients. In this system, the patient will have control over the access to their medical records able to detect medical errors mainly Adverse Event.

During the recent global urgency, scientists, clinicians, and healthcare experts around the globe keep on searching for a new technology to support in tackling the Covid-19 pandemic. The evidence of Machine Learning (ML) application on the previous epidemic encourage researchers by giving a new angle to fight against the novel Coronavirus outbreak. We propose an RGS–SVM model which combines the Randomized Grid Search (RGS) and Support Vector Machines (SVM) to predict which patients have a tendency to be attacked by coronavirus basing on blood indicators. Finally, we proved that the proposed algorithm is capable of identifying persons which are have a tendency for being tested positive for covid-19 with an Accuracy of 87.90%.

Ontologies have become a standard for knowledge representation, supporting the organization and acquisition of knowledge across different domains. However, there are similarities between ontologies and databases in terms of organizing and structuring information, managing relationships between data elements, and supporting querying capabilities. In this paper, our objective is to leverage the information stored in ontologies to automatically construct a database. We propose a novel approach that automates and optimizes the storage of data in a database based on the ontology's structure. We define a set of rules to analyze the ontology, identify concepts, attributes, and relationships, and generate the corresponding database schema. This automation streamlines the database creation process, ensures consistency between the ontology and the database, and enables efficient storage, retrieval, and management of data based on the semantic structure provided by the ontology. To evaluate our approach, we develop a framework that tests the rules using various medical ontologies and generates corresponding databases. We conduct a series of tests to analyze the resulting database and compare it with existing approaches in the literature. Our framework demonstrates effectiveness in terms of generating time and database size.

Internet of Things (IoT) is a growing technology that widely applied to interconnect available medical resources and provide reliable, effective and smart healthcare service to patients. The quality of healthcare is measured by the potential of occurrence of medical errors, cost shrinkage and high speed. This paper proposes a new system to prevent Adverse Drug Events (ADEs) which is a kind medical errors very propagated in health facilities. The proposed system focus on ADE in prescription stage called Prescription ADE system (PrescADE). It improves patient safety, reduces ADE and optimizes the healthcare process. It based on ADE ontology, on IoT and a set of rules to analyze prescription and help doctors. For experimentations, a platform is developed to test rules using 20 patients and three doctors. We compared our proposed system to Computerized Prescriber Order Entry system (CPOE) which is a system to detect ADE of prescription and widely used in hospitals. The results show that PrescADE system has better performance than CPOE system in terms of prescription time and ADE detected. PrescADE detects 314% of ADEs higher than CPOE and 82% of prescription time lower than CPOE.

The enrichment of databases is fundamental to maintain them, as well as the consistency and accuracy of the data. The database becomes useless if it is not up to date. Since there are a large number of databases, an automatic enrichment approach is required. However, until now no efficient approach has been provided in order to cope with this problem. In this paper, we propose a new approach to automate the enrichment of databases. It is based on an ontology, which model domains through sets of concepts and semantic relationships established between them. The proposed approach presents a set of rules to analyze ontologies and databases components and filter subsequently the necessary ones for the database enrichment of databases. We applied our approach in the medical domain that is a renewable domain. Also, it is characterized by a large number of databases and ontologies, and a large volume of data. For experimentations, a platform is developed to test rules using medical databases and medical ontologies. As a result we obtain enriched databases with new components that are either tables, attributes, or records.