Voice signals are crucial for Parkinson’s Disease (PD) classification, as vocal symptoms are common and appear in the early stages of the disease. However, the high dimensionality of datasets can decrease classifier performance, emphasizing the necessity for feature selection tasks. This study aims to implement the Adaptive Hybrid Mutated Differential Evolution (A-HMDE) algorithm for feature selection in the largest public dataset for voice-based PD classification, utilizing the k-Nearest Neighbors (kNN), Support Vector Machine (SVM), Random Forest (RF), and Multi-Layer Perceptron (MLP) classifiers. We observed that A-HMDE performs equally or better than state-of-the-art methods, having selected, on average, 75.43 attributes with an accuracy of up to 88.05% with the RF classifier.