Multinode Cluster Analysis for Prediction of Heart Disease Using Biglm Algorithm

Abstract

The manuscript proposes a novel approach to uncover pattern from a huge chunk of data with minimal time consumption. The goal of the manuscript is to build a robust predictive model to handle big data in patient data. The manuscript follows a master-slave  approach. The training data is divided into n number of chunks and processed. The data involves 32,96,168 row vectors and each row vector holds 14 attributes. The model is given a row vector with 14 attributes whether the system could predict the target class by linear modeling. From the experimentation done in single master and four slave setup, we infer that the system could fit a linear model in 1800 seconds. A comparative study has been made on training the huge chunk of data in a standalone device and in master-slave devices. From the experimental analysis, we infer that the training time of the proposed approach has been reduced by more than half since the proposed approach divides the huge chunk of data into independent chunks and computation is done at multiple nodes.