Feature ensemble learning based on sparse autoencoders for diagnosis of Parkinson’s disease

Abstract

Parkinson’s disease detection through proper representation of the vocal and speech datasets remains an important classification problem. For this problem, we proposed a feature ensemble learning method based on sparse autoencoders. The dataset for this purpose was obtained from UCI, an online repository of comprehensive datasets. Some simulations were conducted over the UCI dataset to confirm the effectiveness of the proposed model. In this paper, the outcomes of the experimentation are compared with the outcomes of stacked sparse Autoencoders and softmax classifier based deep neural network and many classification techniques. Our proposed method yields superior results than DNN. With the proposed model, we obtained a true promising accuracy more than 90%. The outcome of the study also proves that the Feature ensemble learning based on sparse autoencoders method is comparable to other methods present in the literature. The experimental results and statistical analyses are pointing out that the proposed classifier is really useful and practical model for Parkinson’s disease investigation.