Knee osteoarthritis (KOA) stands as a major alignment in musculoskeletal health, affecting millions of people globally. Characterized by the progressive degradation of cartilage within the knee joint, KOA leads to debilitating pain and compromised mobility, significantly impacting the quality of life for those afflicted. Traditional methods for KOA detection and severity grading suffer from subjectivity and data imbalance among severity classes, presenting significant challenges. Leveraging deep learning models, namely, InceptionResNetV2 and NASNetMobile, we propose an automated approach to analyze X-ray images for KOA classification into KL grading. InceptionResNetV2’s deep architecture excels in discerning subtle differences indicative of KOA severity levels, while NASNetMobile offers efficiency in classification tasks, particularly suited to the dataset’s characteristics. Despite their strengths, both models face challenges in distinguishing between specific severity grades, particularly KL1 (doubtful), due to the similarity between classes. Following meticulous evaluation, both models demonstrate promising performance metrics, including precision, recall, and F1 score. InceptionResNetV2 achieves a final accuracy of 65%, while NASNetMobile reaches 63%. The utilization of these advanced machine learning models holds significant potential in revolutionizing KOA classification, offering precise analysis of medical imaging data, and assisting healthcare professionals in patient care and management.