Parkinson’s disease is a progressive neurodegenerative disorder that requires early diagnosis for effective management and improved patient outcomes. The traditional assessment meth-ods are insensitive for early detection of the disease, which creates the space for development of accurate diagnostic techniques. This work presents the development and optimization of a deep neural network multilayer perceptron model for classifying Parkinson’s disease using vo-cal features obtained from the University of California, Irvine, Machine Learning Repository. The dataset contains 22 speech-related attributes that capture motor deficits commonly ob-served in individuals with Parkinson’s disease. The proposed framework consists of an input layer with 22 neurons, three hidden layers, and out of these three two hidden layers are each with 150 neurons and a third hidden layer with 50 neurons, and in the last a single-unit output layer, utilizing ReLU and sigmoid activation functions. To ensure reliability and strong generalization capability, the model was evaluated using a grid search strategy combined with 5-fold cross-validation. This validation scheme supports the development of a practical, non-invasive diagnostic aid for clinical use. The model demonstrated excellent performance, achieving an accuracy of 0.949, precision of 0.973, recall of 0.973, an F1-score of 0.973, and a specificity of 0.50. The results shown that the proposed model is most efficient over conventional classifiers such as support vector machine, random forest, and k-nearest neighbors. This study highlights the effectiveness of a deep neural network–based multilayer perceptron in enabling early and reliable identification of Parkinson’s disease, indicating its value as a supportive instrument for clinical diagnosis.