Akademik Veri Yönetim Sistemi
Medicine Science, cilt.14, sa.2, ss.355-361, 2025 (Hakemli Dergi)
Parkinson's disease (PD) is a neurodegenerative disorder characterized by the gradual degradation of dopaminergic neurons in the substantia nigra pars compacta. The condition presents motor symptoms (such as tremors, rigidity, and bradykinesia) alongside non-motor symptoms (including cognitive impairments and sleep disturbances). Early diagnosis of PD is challenging since symptoms typically appear at advanced stages. In this context, novel methods such as voice analysis have gained importance in detecting the disease in its early stages. Alterations in vocal attributes in persons with PD manifest early in the condition and employing machine learning algorithms to assess these vocal qualities can serve as an effective instrument for early detection. This study aims to examine the utilization of speech analysis for the early detection of PD and the feasibility of machine learning algorithms in these assessments. The study utilizes a PD speech dataset sourced from the UCI Machine Learning repository, comprising 195 voice recordings from 31 individuals, 23 of whom are diagnosed with PD. The study's principal objective is to distinguish between healthy persons and those with PD based on vocal attributes. The associative classification approach was utilized to derive associations between voice characteristics as "If-X then Y" rules. The Ameva algorithm was used to generate the rules evaluated based on support and confidence metrics, selecting the most significant ones for classification. The classification algorithm achieved an accuracy of 92.8%, a sensitivity of 90.5%, and a specificity of 100%. These results indicate that certain voice features, such as fundamental frequency, shimmer, spread, and PPE, are strong indicators for diagnosing PD. The association rules, which showed a perfect confidence level, highlight the importance of these voice parameters in accurately identifying individuals with PD. This study demonstrates the potential of voice analysis, using associative classification, for the early diagnosis of PD. The model showed high accuracy and sensitivity, making voice features like fundamental frequency and shimmer effective indicators of PD. The use of “if-then” rules enhances the interpretability and practical application of the method. Overall, this approach could significantly aid in the early detection of Parkinson’s, improving patient outcomes through timely interventions.