2026-05-13T00:12:34.080Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:11412 2025-05-15T23:37:31Z 3

Augmentation and Substitution of Medical Training Data with Generative Adversarial Networks for Machine Learning, Chapter in CCIS Communications in Computer and Information Science: MDIS 2024: Modelling and Development of Intelligent Systems, Springer, volume 2486 2025 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/11412 https://link.springer.com/chapter/10.1007/978-3-031-87386-7_10 B. Radomirovic L. Jovanovic N. Budimirovic M. Zivkovic N. Bacanin M. Dobrojevic This research underscores the pivotal role of AI in addressing intricate problems, especially in medical diagnosis and predicting treatment outcomes. Despite challenges in algorithm selection, hyperparameter tuning, and limited medical data, simulations with real-world data affirm the effectiveness of machine learning, notably showcasing the random forest model with an impressive 85.19% accuracy. A significant aspect of this work involves the exploration of Generative Adversarial Networks (GANs) for data augmentation and synthesis. GANs enhance classifiers like MLP and AdaBoost but present challenges for decision tree and KNeighbors models. Additionally, leveraging fully synthetic data for training proves promising, offering a potential solution to data scarcity. Feature importance analysis emphasizes the impact of treatment frequency on patient outcomes, enhancing model interpretability. In conclusion, this research addresses challenges, introduces novel GAN-based approaches, and provides valuable insights to advance practical AI applications, particularly in effective data augmentation and synthesis using GANs, ultimately improving the prediction of treatment outcomes. bookPart Springer, Cham 138 152 10.1007/978-3-031-87386-7_10 CCIS Communications in Computer and Information Science: MDIS 2024: Proceedings of International Conference on Modelling and Development of Intelligent Systems, volume 2486