Parkinson’s Disease (PD) remains a challenging neurological condition for all the healthcare professionals and researchers. There exists no cure for the illness but if diagnosed at an early stage, with personalized line of treatment the disease can be controlled and cured up to some extent. Artificial Intelligence (AI) is considered as a powerful and potential tool for providing with precise results for early diagnosis of PD. Different types of datasets including voice samples, hand drawings, clinical data and medical images serve as a foundation for training different machine learning and deep learning models for making predictions and giving insights. Different deep learning architectures are being used over a decade for giving considerable results, but the transfer learning techniques have proven out to outperform others giving the most optimal solutions. These models have high adaptability but training them on high resolution image data requires significant amount of time and resources. To address this, different hardware acceleration techniques have been used to reduce the training time and optimize the process. The dataset consists of hand drawn spiral images, consisting of Healthy and Parkinson classes, which are used to train and test a proposed hybrid transfer learning model (H-TL) which gives an accuracy of 96.28% with a loss of 0.113 only. The model is tested over the dataset by fine-tuning and hyper parameter tuning, and it’s observed that the proposed model’s accuracy shows a considerable improvement while a remarkable decrement in loss of the model after tuning the hyper-parameters and fine tuning. The research work also examined and analyzed the computational complexity of the proposed model.