A novel hybrid machine learning approach for early prediction of Parkinson’s disease severity using optimized feature selection and ensemble learning