Parkinson’s disease is a progressive nerve cell loss condition that affects millions of individuals globally, both motor and non-motor functions. Early and precise prediction of the severity of the disease is critical for timely intervention and individualized treatment. In this research, we introduce an ensemble learning method that integrates eight machine learning models, such as linear regression, support vector regression (SVR), random forest, gradient boosting, multilayer perceptron (MLP), decision tree, K-nearest neighbors (KNN), and AdaBoost, to forecast the severity of Parkinson’s disease based on a detailed dataset of speech features. The performance of the models was tested with primary metrics like mean squared error (MSE), mean absolute error (MAE), R-squared (R2), and mean absolute percentage error (MAPE). The outcome demonstrated that Combo 4 combination of random forest, SVR, MLP, and linear regression performed better than any other model combination. It had the smallest MSE (6.9794 for motor UPDRS and 10.7442 for total UPDRS), the smallest MAE (1.8732 for motor UPDRS and 2.3339 for total UPDRS), and the largest R2 values (0.8907 for motor UPDRS and 0.9030 for total UPDRS). The results of the study reveal the efficiency of the ensemble learning method in determining the severity of Parkinson’s disease with high precision. The research shows the potential of employing noninvasive speech data analysis as a diagnostic tool for early diagnosis and enhanced clinical management of Parkinson’s disease.