PPMI Research

The main goal is to create a system in order to classify in a quick way healthy people from people suffering from Parkinson's disease. We have developed a process based on Latent Semantic Analysis, which finds hidden meanings in a text. For instance ignoring synonims. We have extended the process using a neural network technique text2vec that gives a numerical representation on the documents and finds the most similar. Both technologies produce a similiarity matrix (cosine distance). Then we use clustering algorithms to classify the patients. We've find out that the most accurate techniques, benchmarked using F-measure, were text2vec with K-means and LSA with Fuzzy. Proposals for further development could be a better training of the neural network text2vec.

For further information check the PowerPoint in this repo

Google Drive