Investigation on optimizing analytics using metaheuristic algorithms

Abstract

Data is produced from various devices like sensors, mobile devices, IoT, cyber physical systems and text, pictures, videos through social networks, web and mobile applications. Processing and analysing such large volumes of data using learning algorithms to produce optimal insights for enhancing decision making is often termed as Analytics. The analytics model designed using conventional machine learning algorithm pose challenges in maximizing accuracy as the algorithms are based on parameters essential to build the model. Also, when the model is fed with too many irrelevant features in the dataset, the performance of model built using Machine Learning (ML) may be degraded. Some other challenges that the traditional machine learning algorithm incurs because of random parameters and vast number of irrelevant features is maximum error rate and huge computation time. The research work aims for optimizing analytics by integrating metaheuristic algorithm with ML algorithm and its performance is investigated. newlineThe proposed System for Accelerating and Optimizing Analytics (SAOA) includes building a predictor model by estimating the parameter of ML algorithm, designing feature selection algorithm with the focus to improve further and designing a system to choose optimal analytic service provider. Initially, Improved Radial Basis Function Neural Network (IRBFNN) is proposed using Particle Swarm Optimization integrated with K-Means to choose optimal cluster centers for the hidden layer newline