Boosted Autoencoder Decoder Network for Parkinson's Disease Detection with Feature Selection

Abstract

Parkinson's is a major brain disease that is most common in people who have stressed lives or bustling daily schedules. There are times when this problem ends in death. With 755 characteristics, many experts made a system using machine learning that can tell early on if someone has a sickness and save their life. Like other illnesses, it's hard to tell how the disease will grow or show up because there aren't many signs that are the same for everyone and some symptoms can be different from person to in person. You can tell someone has Parkinson's disease by looking at their motor and non-motor signs. This essay looked at the features that were taken from the recorded sound signals. Image processing methods are expensive ways to look at audio data, but the machine requires to work on recordings in order to work well. In this way, the suggested system looked at the data format's features. The model got rid of the features that weren't needed by combining an improved auto encoder in a gradient boosting method to define the exact hierarchies signs and happenings. Researchers used statistical methods, ML techniques, and more, such as Recursive Feature Elimination and Correlation analysis. The measurements used to judge these methods don't work well because the data has a lot of dimensions. With a 95.18% success rate, the technique uses the suggested method to cut the number of traits down to 94. We change the current autoencoder's design so that it works with the data we have.