Parkinson’s disease detection using existing machine learning algorithms

Sammanfattning

Background. Parkinson’s disease (PD) is a neurodegenerative disorder characterized by symptoms like tremors, stiffness, slow movement, and impaired balance this disease is caused by the small amount of dopamine generated in the brain. Early detection of this disease can save the patients, while treatments only focus on the symptoms. Even though it is a challenges with PD can lead fulfilling lives with proper management. Many of the research are ongoing to improve early diagnosis and personalized treatments for this complex neurological disorder. In terms of medical and cutting-edge technologies, many researchers are ongoing. Objectives. The main goal of the thesis is to model and train the selected machine learning algorithm. It involves different stages like collecting data, preprocessing, splitting, and evaluating the model. Based on the accuracy and f1 scores of the algorithms, a better algorithm is selected for detection. Methods. Experimentation methodology is used to answer the research question the experimentation involves steps like collecting and preprocessing data from Kaggle, splitting it into training and testing sets and 5 fold cross validation, training the models, and testing the model on the selected algorithms Artificial Neural Networks (ANN), Light GBM, and XGBoost. From the results obtained in accuracy and f1 score, the better algorithm is justified which help in the early detection of disease. Results. After the experimentation and calculating the metrics it is identified that Light GBM is to be a better algorithm for predicting the PD and it attained a testing accuracy of 95.6% and also attained a better F1 score which is 0.96. Upon comparison with ANN, XGBoost, Light GBM. Light GBM is considered as the better algorithm. Conclusions. The methodology ensured thorough categorization of Parkinson’s disease using Kaggle data, balanced with SMOTE to address class imbalance. After meticulous cleaning and normalization, Light GBM, XGBoost, and ANN were trained and evaluated for accuracy and F1 scores. Light GBM emerged as the optimal choice, showcasing the highest accuracy (95.6%) and an outstanding F1 score of 0.96, indicating balanced recall and precision. Its superior performance in accurately identifying patients with Parkinson’s disease positions Light GBM as the preferred algorithm for this task over XGBoost, which achieved a lower F1 score of 0.92.