جامعة تكنولوجيا المعلومات والاتصالات

This project addresses a significant problem in the field of intelligent medical systems and clinical decision support: the lack of predictive precision, data balance, and clinical transparency in early obesity risk forecasting. The proposed solution is based on an intelligent health informatics pipeline that couples high-performance machine learning algorithms with Explainable AI (XAI) using TreeSHAP. The system aims to predict and categorize individuals into seven distinct weight and obesity risk tiers using 16 multi-factorial lifestyle, behavioral, and demographic parameters rather than relying solely on the static Body Mass Index (BMI). Ten machine learning models were developed and evaluated using Stratified 10-Fold Cross-Validation and SMOTE to eliminate structural dataset imbalances. The expected outcomes include a highly robust predictive engine; the LightGBM model achieved a peak standalone accuracy of 97.54%, while a custom Soft Voting Ensemble framework was deployed at 96.00% accuracy due to its superior architectural stability. Notably, the system achieved a flawless 100% precision and recall for high-risk Obesity Type III cases, ensuring zero false negatives. Furthermore, a responsive, interactive Streamlit web application was developed and deployed, featuring an intuitive clinical user interface that provides healthcare practitioners with real-time, transparent risk classifications and personalized local SHAP force plots for proactive lifestyle interventions.

This research addresses two critical challenges in smart medical systems: fragmented health data with cybersecurity vulnerabilities, and inefficient medical inventory management. The proposed solution, Health-Eco Hub, is a unified ecosystem that integrates AI-driven predictive analytics with a real-time Intrusion Detection System (IDS). In addition, the project incorporates MedStock, an intelligent inventory management module that applies a weighted Risk Scoring Model and a Simple Moving Average (SMA) algorithm for demand forecasting. MedStock also introduces role-based dashboards (Manager, Staff, Researcher) to ensure accountability and operational transparency.The implementation results demonstrate significant improvements: a 95% reduction in cybersecurity risks, 92% accuracy in predicting inventory shortages, an 80% improvement in medical record and inventory retrieval speed, and alert latency under 100ms. Collectively, these outcomes transform static health and inventory records into actionable, life-saving insights, enhancing patient safety, institutional efficiency, and digital resilience.

This project proposes an intelligent predictive maintenance framework for industrial machinery and demonstrates its effectiveness through a case study based on agricultural equipment operating in demanding environmental conditions. The framework combines Internet of Things (IoT) sensing technologies with machine learning to enable condition-based maintenance and the early detection of equipment faults. An Arduino-based prototype equipped with temperature, vibration, and engine RPM sensors was developed to continuously monitor key operating parameters. The collected sensor data, together with an open-source dataset, were analyzed using a Random Forest classification model to identify potential faults before they progressed into critical failures. To improve system usability, the framework incorporates an intuitive diagnostic interface that converts complex analytical results into clear and practical maintenance recommendations, making the system accessible even to users with limited technical expertise. Experimental evaluation demonstrated that the proposed framework achieved an overall fault detection accuracy of 92.11%. The results indicate that the system can effectively support predictive maintenance by reducing maintenance costs, minimizing unplanned downtime, extending equipment service life, and improving operational efficiency. Although the framework was validated using agricultural machinery as the primary case study, its architecture is sufficiently flexible to be adapted for a wide range of industrial machinery and predictive maintenance applications.

More than 3,000 preventable deaths occur on Iraqi roads each year, partly due to the lack of integration between road hazard detection, geospatial mapping, and emergency response systems. Existing CCTV networks, citizen-reporting applications, and standalone machine-learning solutions operate independently, limiting their effectiveness in supporting timely decision-making. This paper presents RASID (Road AI Surveillance Intelligence Detection), an integrated MLOps and GIS platform for real-time road hazard monitoring and fleet intelligence. The proposed system streams image frames and GPS coordinates from fleet vehicles to a centralized server, where a YOLO11s detection model performs hazard detection and stores validated events in a PostGIS database as geotagged records. Detected events are distributed through a WebSocket-based GIS platform, enabling real-time notifications for drivers, maintenance teams, and emergency responders. In addition, an active-learning pipeline supports continuous model improvement through automated retraining and deployment. A pilot deployment involving ten vehicles achieved 78.0% [email protected], 74.7% precision, and 81.1% recall, while sustaining up to 120 frames per second of server throughput and maintaining sub-second event dissemination latency. The results demonstrate the feasibility of combining computer vision, geospatial intelligence, and MLOps into a unified platform for intelligent road safety management.

This project addresses a significant problem in the field of smart healthcare and IoT-based medical monitoring systems, particularly the early detection of epileptic seizures and continuous monitoring of patients’ vital signs in different environments such as home, school, work, and public places. The proposed solution is based on a wearable intelligent system that integrates physiological sensors such as MAX30102 for heart rate and blood oxygen saturation monitoring, and MPU6050 for motion and fall detection, using ESP32 as the main controller, along with machine learning techniques for data analysis and seizure prediction, and a mobile application for real-time alerts. The aim of this system is to improve early seizure detection, reduce emergency response time, and enhance patient safety both indoors and outdoors while maintaining privacy and reducing social embarrassment. The expected outcomes include a reliable and accurate seizure and fall detection system, faster emergency response, and continuous smart monitoring that improves the overall safety and quality of life for patients in daily activities. يتناول هذا المشروع مشكلةً جوهريةً في مجال الرعاية الصحية الذكية وأنظمة المراقبة الطبية المعتمدة على إنترنت الأشياء (IoT)، والمتمثلة في صعوبة الكشف المبكر عن نوبات الصرع ومتابعة المؤشرات الحيوية للمرضى في مختلف البيئات مثل المنزل والمدرسة والعمل والأماكن العامة. ويستند الحل المقترح إلى نظام ذكي قابل للارتداء يعتمد على حساسات حيوية مثل MAX30102 لقياس نبض القلب وتشبع الأكسجين، وMPU6050 لرصد الحركة والسقوط، مع استخدام ESP32 كوحدة تحكم رئيسية، وتقنيات التعلم الآلي لتحليل البيانات والتنبؤ بالنوبات، وربطه بتطبيق موبايل لإرسال التنبيهات الفورية. ويهدف هذا النظام إلى تحسين الكشف المبكر عن نوبات الصرع، وتقليل زمن الاستجابة للحالات الطارئة، وتعزيز سلامة المريض وحمايته أثناء وجوده في المنزل أو خارجه مع الحفاظ على خصوصيته وتقليل الإحراج الاجتماعي. وتشمل النتائج المتوقعة تحقيق نظام دقيق وموثوق للكشف عن النوبات والسقوط، وتحسين سرعة الاستجابة للطوارئ، وتوفير مراقبة ذكية ومستمرة تدعم سلامة المرضى في حياتهم اليومية.

نظام " Smart Vision "، وهو نظام ذكي مخصص للكشف عن الأسلحة وتتبعها في الوقت الفعلي، يعتمد على تقنيات التعلم العميق والرؤية الحاسوبية. يستخدم النظام نموذج " Ultralytics YOLO26m "، الذي يُعد أحدث إطار عمل موحّد للكشف عن الكائنات في الوقت الفعلي، ويتميز بقدرة أصلية على الاستدلال المباشر ) end-to-end inference ( وبنية معمارية لا تعتمد على خوارزمية كبت التكرارات غير القصوى ) NMS-free (. وقد استُخدمت مجموعة بيانات مخصصة تضم 20,000 صورة جُمعت ذاتيا،ً قُسّمت إلى: 80 % للتدريب، و 10 % للتحقق من الصحة ) validation ،) و 10 % للاختبار. حقق النموذج دقة ) Precision ( بلغت 93.7 %، ومعدل استدعاء ) Recall ( بنسبة 89 %، ومتوسط دقة متوسط ) mAP@50 ( بنسبة 94 %، بالإضافة إلى قيمة mAP@50-95 بلغت 83 %، مما يبرهن على دقة عالية في رصد الأسلحة في ظل ظروف بيئية متنوعة. ويُقدم النظام المقترح حلاً عمليا وقابلاً للتطبيق الفعلي لتعزيز السلامة العامة من خلال أنظمة المراقبة بالفيديو الآلية.

Sunbula.ai، مساعد زراعي مبني على الذكاء الاصطناعي يعمل بالكامل على الهاتف المحمول دون الحاجة إلى الاتصال بالإنترنت. حيين يلتقط المزارع صورة للمحصول او النبات المتضرر أو ممكن ان يكتب الى المساعد باللغة العربية واللهجة العراقية، فيعالج النظام المدخلات مباشرة سواء كانت صورة او كتابة على الجهاز محلياً ويقدم استجابة نصية باللغة العربية العراقية الملائمة، من الخدمات التي يمكن ان يقدمها المساعد هي توصيات وارشادات مخصصة للمزارع تعتمد على نوع النبات والتربة والظروف المناخية المحلية. يساهم سُنبلة المساعد الذكي على إزالة الصعوبات والعوائق التي تواجه المزارعين في الوصول إلى الإرشاد المتخصص في المجتمعات الريفية، خاصة في بيئة تعاني من ضعف خدمات الاتصالات والانترنت، ومحدودية القراءة والكتابة، والبعد الجغرافي.