MAPREDUCE BASED BIG DATA FRAMEWORK USING DEEP BELIEF NONLINEAR EXPONENTIAL CLASSIFIER FOR DIABETIC DISEASE PREDICTION

Abstract

Healthcare domain is a very distinguished research area with swift technological evolution and surging data progressively. With the intent of extensive healthcare data Big Data Analytics is turning up to be an emerging viewpoint in Healthcare domain. Millions of patients look for treatments globally with numerous procedures. Deep Learning (DL) is an encouraging mechanism that aids in early disease diagnosis and could be beneficial for the practitioners in decision making. This paper aims at building a Deep Learning and MapReduce based Big Data method called, Non-linear Auto Correlated Encoding and Normalized Exponential Classification (NACE-NEC) for diabetes prediction. In order to predict it more accurately, this paper proposes a diabetic disease prediction model that combines MapReduce pre-processing, correlated feature selection and classification. Firstly, the diabetic prediction dataset is pre-processed using Batch Normalized Covariate Transpose Propagated MapReduce. Then, combined with two factor correlation analysis between features using correlation coefficient function based on Non-linear Auto Encoding is performed with the optimal feature subset as the feature input. Finally, the Normalized Exponential Classification is used to make robust differentiation between diabetic and non-diabetic via cross entropy as loss function. To evaluate the NACE-NEC methods performance, five different performance metrics, disease prediction time, misclassification rate, precision, recall and accurate rare validated and analyzed. The NACE-NEC achieved higher performance compared to other state-of-the-art methods on our collected diabetic prediction dataset demonstrating the efficiency of the method in reducing misclassification rate by 40% while improving overall accuracy by 22% and precision extensively.