Parkinson’s disease (PD) presents a significant challenge due to its slow progression and symptom overlap with other conditions, making early detection crucial for effective treatment. Common symptoms include tremors, muscle rigidity, slowed movements, and balance issues, alongside psychiatric symptoms. Various machine learning approaches have been explored to aid in early PD detection. One prevalent method involves analyzing handwritten spirals, which can reveal distinct patterns indicative of Parkinson’s. However, traditional feature extraction techniques often yield subpar accuracy, which is unacceptable given the severity of the disease. To address this, we propose a novel deep learning model designed to efficiently extract optimal features for improved accuracy. Our model utilizes a predictive approach, leveraging a dataset of spiral drawings in PD to assess the severity. By selecting the most relevant features, our model aims to achieve higher performance accuracy, facilitating earlier detection and intervention.