A Parallel and Distributed Radial Basis Function Network for Big Data Analytics

Radial Basis Function Network (RBFN) is one of the versatile neural network architectures, in that it can be used to solve regression as well as classification tasks. The proposed work implements RBFN in a parallel and distributed environment of Apache Spark. The new version of RBFN is referred to as PRBFN. The main contribution in this work is embedding of K-Means|| (a parallel distributed clustering method) in between input and hidden layer and implementation of parallel least square estimation employing outer product of matrices. The effectiveness of the PRBFN was tested on HEPMASS, HIGGS, ccFraud, and Amazon Movie Review (for sentiment classification) datasets under 10 Fold Cross Validation (10-FCV) setup. It is emphasized that the chief objective is to present a distributed and parallel version of the RBFN to handle big datasets.