Computational Intelligence Based Modelling of Polyneuropathy Diagnosis

This study addresses the complexity of diagnosing Polyneuropathy (PNP), a group of neurological disorders affecting peripheral nerves. The research introduces a hybrid medical data analysis framework, combining machine learning with computational intelligence methods, to enhance the diagnostic process. The proposed methodology involves data preprocessing, case classification, model evaluation, and knowledge extraction. The study uses data from two German University Hospitals (Jena and Tübingen) to define two classification problems and employs the C5.0 algorithm optimized with AdaBoost. The introduced data balancing method (LCC—Leveling of Cases per Class) significantly outperforms the SMOTE method, achieving an average classification accuracy exceeding 95%. The study presents classification rules in medical knowledge mining, emphasizing the importance of Nerve Conduction Studies and Echo Intensity in diagnosing specific PNP types. The work contributes to data analysis and medical knowledge, providing valuable insights for informed decision-making in the polyneuropathy diagnostic process. Future research avenues include refining data preprocessing and exploring the cost implications of medical decisions guided by the proposed framework.