A neurological disorder called Parkinson's disease (PD) can result in a variety of motor and non- motor symptoms. Bradykinesia, tremors, postural instability, irregular gait, weariness, depression, and sleep difficulties are some of these symptoms. For disease treatment to be successful, early and precise diagnosis is essential, particularly in the healthcare industry. Convolutional neural networks (CNNs) were utilized in this study to diagnose Parkinson's disease, and CNN- long short-term memory networks with the Hoehn and Yahr rating scales were employed to predict severity ratings. We used 93 PD patients and 73 healthy subjects’ gait patterns from our dataset, which we downloaded from Physio Net. Three walking tests were used to categorize the dataset: Performing two tasks at once, rhythmic Auditory Stimulation and Treadmill walking. Eight Force Sensitive Resistors under the foot, which recorded the Vertical Ground Reaction Forces, allowed us to record the gait patterns. We applied our proposed classifier to extract spatiotemporal variables, such as the swing phase, stance phase, and gait phase (a mix of swing and stance phase). We further divided the gait phase into toe-off, heel-off, heel-strike, midstance, and foot flat. To predict gait abnormalities, we employed major temporal features such as stride time, step time, stance time, and swing time, together with spatial features like stride length, step length, gait speed, and cadence. During the training phase, we employed deep learning to implement the feature extractor, which is a more effective method than manual implementation. We employed CNN for our study..