Benrezkallah ,Kawther2025-11-032025-11-032025-07-03https://dspace.univ-tlemcen.dz/handle/112/25168Radiation therapy (RT) playing an important role in cancer treatment modality that using high-energy radiation to killing tumors without damaging healthy tissues (OARs). However, still achieving optimal beams radiation delivery accuracy remains a significant challenge. The advanced techniques such as VMAT, IMRT, IGRT have enhanced treatment precision , but still need improving incorporating new technologies . Artificial intelligence, particularly ML and DL offers potential to improving revolution in RT. One of possible way is modeling behaviors of radiation beams. AI techniques enable to analysis of massive datasets and identify complex patterns that are difficult to detect by traditional methods.This thesis explores the application of AI in a part of modeling behaviors of radiation beams which is classifying OARs into categories. The research focusing on using clinical data from real cases ― cavum cancer ― about 10 patients with one to three plans to develop an AI model for classifying OARs into : safe , medium risk , high risk , unknown ( this for clinical volume which it varies from one plan to another ) . The developed model using dose metrics such as maximum, mean, and median doses, along with anatomical structure types and other clinical variables to assess OAR risk levels. A CatBoost Classifier was employed and demonstrated excellent performance in terms of accuracy, recall, precision, and F1 score, ensuring reliable identification of high-risk cases—an essential requirement in medical decision-making. Data preprocessing steps included handling class imbalance using Random Over Sampling, followed by feature selection and model training using the PyCaret library. The results indicate that the AI model successfully identifies all high-risk and medium-risk cases, with minimal false positives in the "Safe" category. Feature importance analysis revealed that dose-related parameters were the most influential in determining risk classification. This study demonstrates how AI can support radiation oncologists and medical physicists in making more informed and efficient treatment decisions. By integrating AI into RT planning, this approach enhances the quality and safety of treatment plans, reduces planning time, and improves patient outcomes. Despite limitations related to dataset size and technical expertise, the findings underscore the promising role of AI in transforming radiotherapy workflows and advancing personalized cancer treatment. KEYWORDS: Artificial Intelligence, Machine Learning, Deep Learning, Radiotherapy, Beam Modeling, Organ at Risk (OAR), Dose Prediction, Treatment Planning, CatBoost Classifier, Medical PhysicenModeling behavior of radiation beams in radiotherapy using AIThesis