Parkinson's illness is an incurable nervous disorder marked by a fall in dopamine levels in the brain of humans. It ranks as the second most common neurological disorder. Developing tools for early and automated diagnosis of Parkinson's illness is crucial. This study aims to enhance the advance of computerised methods for detecting Parkinson's disease (PD) utilizing Magnetoencephalography (MEG). MEG sub-bands were created using discrete wavelet transform (DWT). Many features were extracted from sub-band decomposed signals, and the Binary grey wolf optimizer (BGWO) was employed to determine the most significant features. Multiple machine learning models were employed, utilizing these crucial properties as input. The proposed methodology is assessed using data obtained from The Swedish National Facility for Magnetoencephalography Parkinson’s disease dataset (NatMEG-PD), which comprises a data sample that contains 66 individuals with Parkinson’s illness and 68 individuals who do not have the illness. The Random Forest classifier produces accuracy, specificity, sensitivity, precision, and F-Score of 99.46%, 99.60%, 99.30%, 99.53%, and 99.41, respectively. The proposed methodology would benefit neurologists’ diagnostic methods and clinical performance.