Deep Learning for prediction of Amyotrophic Lateral Sclerosis using Stacked Auto Encoders

Healthcare is an emerging area in big data. Raw data contains lot of noise in it, hence cannot produce good results when processed. There is a need to improve the quality of data. This study shows how the prediction accuracy can be improved if the quality of data is improved. Previous work on issues related to variety and veracity have already been cited. Here the issues related to prediction are addressed. The dataset contains 1,047,253 records of patients having amyotrophic lateral sclerosis (ALS). Missing data values are filled and later used for prediction. Predicting the progression of the disease was calculated using stacked auto encoders. The results were compared with traditional techniques like random forest and support vector machine. A similar study was conducted using random forests and the accuracy obtained was only 66%. This paper presents a study on how to predict the progression of ALS using deep learning and an accuracy of 88% was achieved which is far more than the accuracy obtained on raw data. The study thus demonstrates the fact that accuracy increases with better data.