https://dspace.univ-guelma.dz/jspui/handle/123456789/674 Fri, 16 Oct 2026 15:19:00 GMT 2026-10-16T15:19:00Z https://dspace.univ-guelma.dz/jspui/handle/123456789/19542 Titre: Un Système de recommandation explicable pour le suivi thérapeutique personnalisé des patients diabétiques Auteur(s): BOUZAROURA Sirine, SAMSAR Ahmed Amine Résumé: Insulin-treated diabetes requires regular monitoring, as fluctuations in blood glucose levels can lead to hypoglycemic or hyperglycemic episodes that are difficult to anticipate. With the advent of continuous glucose monitors and connected devices, patients and healthcare professionals now have access to more data, but its interpretation remains complex. This dissertation presents "GlucoAI," a prototype glycemic monitoring support system based on data from the Internet of Medical Things. The aim of this work is to design and implement a solution capable of organizing blood glucose data, estimating short-term glycemic risk, explaining the predictions produced by artificial intelligence models, and providing personalized recommendations that are interpretable by both the patient and the clinician. The proposed solution is built on an architecture comprising a FastAPI backend, a PostgreSQL database, a mobile application developed with Expo/React Native, and a web presentation portal. The experimental evaluation was conducted on a controlled synthetic dataset comprising 11 patient profiles over 14 days. Four prediction tasks were studied: hypo-30, hypo-60, hyper-30, and hyper-60. The selected models — notably EBM, Random Forest, and XGBoost — achieved high results on this experimental dataset, with AUROC scores ranging from 0.9934 to 0.9995. The system also incorporates an explainability layer based on SHAP, LIME, and native EBM, as well as a clinician-validated recommendation mechanism. The results demonstrate the technical feasibility of an explainable glycemic monitoring prototype. However, the performance obtained remains limited to the synthetic setting and will need to be confirmed on real-world data in future work. Thu, 01 Jan 2026 00:00:00 GMT https://dspace.univ-guelma.dz/jspui/handle/123456789/19542 2026-01-01T00:00:00Z https://dspace.univ-guelma.dz/jspui/handle/123456789/19523 Titre: OculiSight: A Citizen-Centered Mobile Eye Health Platform for AI-Assisted Screening, Visual Assessment, and Remote Ophthalmic Care in Non-Clinical Settings Auteur(s): BRAHMIA CHAIMA, ALAIMIA HADIA Résumé: Visual health represents a major public health challenge, particularly in Algeria where access to ophthalmology services remains limited in rural and underserved regions. To address this issue, this work presents OculiSight, an innovative citizen-centered digital health platform that combines artificial intelligence and mobile accessibility for prelimi nary screening of ocular pathologies. The system enables users to capture or import eye images via smartphone and per form automated analysis in both online and offline modes. It generates structured screen ing outputs including disease predictions, activation maps (YOLO-based localization and LayerCAM explainability), associated symptoms, and preliminary recommendations. The platform integrates deep learning models such as ResNet50 and MobileNetV3 for classi f ication, alongside YOLOv8n and YOLOv11n for detection tasks, deployed in a hybrid architecture supporting both server-side inference and on-device processing via Tensor Flow Lite. In addition to AI-based screening, OculiSight incorporates standardized vision tests (visual acuity, color vision, night vision, and Amsler grid assessment), secure patientdoc tor communication, appointment scheduling, geolocation of ophthalmologists, and bidirec tional data synchronization. The system is designed as a non-diagnostic decision-support tool intended for preliminary assessment and early detection support. The platform is validated through expert collaboration and clinical feedback, ensuring relevance to real-world ophthalmology practice. By combining offline capability, AI-driven analysis, and integrated care services, OculiSight aims to bridge the gap between accessi bility and advanced medical screening in resource-constrained environments. Thu, 01 Jan 2026 00:00:00 GMT https://dspace.univ-guelma.dz/jspui/handle/123456789/19523 2026-01-01T00:00:00Z https://dspace.univ-guelma.dz/jspui/handle/123456789/19499 Titre: AI-Based Decision Support Systemfor Autism Spectrum DisorderDetection Using EEG Signals Auteur(s): GUERROUI, Abdelghafour Résumé: Autism Spectrum Disorder (ASD) is a neurodevelopmental condition whose earlydetection remains a major challenge due to the complexity and subjectivity of tra-ditional diagnostic methods.This work presents a novel system, NeuroDetect ASD, an intelligent web-basedplatform that combines electroencephalography (EEG) analysis with Artificial In-telligence (AI) techniques to support the early screening of ASD. The proposedsystem integrates an EEG processing pipeline that includes preprocessing, featureextraction, Principal Component Analysis (PCA), and an ensemble machine learningmodel based on AI techniques for automatic classification.In addition to the AI model, the platform provides secure patient management,automated medical report generation, and a multi-tenant architecture that enablesmultiple healthcare institutions to use the system while ensuring data confidentiality.The obtained results demonstrate the potential of NeuroDetect ASD as a reliableand scalable decision-support solution capable of assisting healthcare professionalsin achieving faster and more objective ASD screening, while promoting the adoptionof intelligent digital technologies in clinical practice. Thu, 01 Jan 2026 00:00:00 GMT https://dspace.univ-guelma.dz/jspui/handle/123456789/19499 2026-01-01T00:00:00Z https://dspace.univ-guelma.dz/jspui/handle/123456789/19492 Titre: Immersify: Enabling University 4.0 through AR/VR-Driven Immersive Learning (“My Smart Campus”) Auteur(s): DJEBALA AMIRA, LABADLIA MALAK Résumé: In the wake of the Fourth Industrial Revolution, higher education institutions are undergoing a fundamental redefinition, transitioning toward the University 4.0 model—an intelligent, data-driven, and accessible academic environment in which artificial intelligence, augmented reality, and real-time technologies coexist to enrich the learning experience. The present thesis addresses this transformation by proposing a comprehensive Smart Campus platform structured around four role-based interfaces. The Normal Student interface integrates immersive 3D campus navigation, QR code-based room identification, real-time display of active modules and teaching staff, and a suite of AI-driven tools encompassing automated PDF summarization, quiz generation, mind map construction, and a conversational chatbot. Content personalization is achieved through a reinforcement learning model employing the Q-Learning algorithm and the Bellman optimality equation to dynamically tailor recommendations to each student’s academic profile. The Visually Impaired Student interface offers an entirely voice-controlled environment from authentication to content consumption, supported by a context-aware voice assistant, spoken quiz delivery, and environmental perception through TensorFlow COCO-SSD object detection. The Teacher interface facilitates the structured publication of multiformat educational resources, while the Administrator interface ensures centralized governance of campus entities with real-time WebSocket-driven schedule Thu, 01 Jan 2026 00:00:00 GMT https://dspace.univ-guelma.dz/jspui/handle/123456789/19492 2026-01-01T00:00:00Z