Parkinson’s disease (PD) is a neuro-degenerative illness triggered by brain cell malfunction causing 60–80% reduction in the production of dopamine, the natural neurotransmitter that regulates movement. Analysis of speech abnormalities forms the basis of the diagnosis of PD as speech is deteriorated in its early stage. The speech has been analysed using machine learning (ML) techniques to identify Parkinson’s patients from those that are normal. This paper presents a new feature extraction method intrinsic mode function instantaneous amplitude deviation (IMFIAD) to efficiently represent the PD speech. The PC-GITA dataset is used with support vector machines (SVM) classifiers to detect the Experimental results show that the proposed technique superior to existing studies in which IMFIAD with SVM achieved highest accuracy of 87% in vowel /i/ and 84% in word /apto/.