تشخيص بيماري پاركينسون با استفاده از الگوريتم‌هاي يادگيري ماشين مبتني بر مهندسي ويژگي و منطق فازي

Parkinson’s disease is one of the most common neurodegenerative disorders, primarily affecting the motor system. Although early diagnosis plays a crucial role in improving patientsʹ quality of life an‎d slowing disease progression, subtle an‎d often imperceptible early symptoms pose significant challenges to conventional diagnostic methods. This study addresses this issue by proposing a comprehensive an‎d interpretable machine learning framework for the early diagnosis of Parkinson’s disease. The proposed framework integrates Ran‎dom Forest, Deep Neural Networks, an‎d a Logistic Regression meta-classifier in a stacked ensemble learning architecture, effectively leveraging the complementary strengths of individual learners to significantly enhance model accuracy an‎d generalizability. The feature engineering pipeline includes statistical analysis, SMOTE technique for class balancing, an‎d feature importance extraction using Ran‎dom Forest an‎d SHAP algorithms. Subsequently, a fuzzy logic system is employed to categorize the extracted features into three levels of clinical importance. Ultimately, the 9 selec‎ted features, derived from this process, are not only statistically significant but also clinically interpretable. These features primarily related to specific aspects of patients’ speech enable clinicians to gain deeper insights into the patient’s condition, monitor disease progression, an‎d make more precise treatment decisions. The proposed model was eva‎luated across 20 ran‎dom train an‎d test splits. The performance metrics achieved include an AUC of 98%, accuracy of 96%, precision of 96%, sensitivity of 99.1%, an‎d an F1-score of 97.4%. These results demonstrate the model’s superior performance compared to conventional classifiers an‎d its alignment with state of the art approaches. Moreover, the model’s high interpretability an‎d computational efficiency make it a promising an‎d practical tool for use in clinical environments, enabling early screening an‎d monitoring of patients in the initial stages of the disease. Overall, the proposed framework goes beyond a mere predictive system an‎d functions as a clinical decision-support tool that enhances the diagnosis an‎d treatment process by offering a more profound understan‎ding of the patient’s condition.

مريم ذكري

جلال ذهبي

جعفر قيصري , مرضيه كمالي