Metaheuristic based feature selection methods and ensemble classification models in cloud network analytics

Abstract

Cloud network monitoring data exhibits dynamicity and newlinedistributiveness. Hence Machine Learning (ML) models are adapted to a newlineparticular dataset can fast tend to be insufficient. However, its accuracy may newlinebe lost after some time owing to variations in input data and their features. newlineTherefore, Feature Selection (FS) has emerged to be an optimal solution to newlinereduce the number of features in the dataset. Distributed learning with newlinedynamic model selection and time-dependent models is often required. newlineEspecially, a dynamic mechanism for auto-selection and auto-tuning of newlinemachine learning models is necessary, which exhibits considerable change newlineover time. The major contribution of the work will focus on developing data newlinemining methods with feature selection, and ensemble learning by ML, and DL newlinemodels for cloud data analytics. Three major contributions have been newlineperformed in this work. newlineThe first contribution of the work,is Cloud Development Q Learner newlineMachine Learning Operations (DevQLMLOps) architecture which introduced newlinefor autotuning and selection based on container orchestration and messaging newlinebetween containers. Adaptive Kernel Firefly Algorithm (AKBA) based newlinedynamic model is proposed for feature selection in Cloud network monitoring newlinedata. It is inspired by the flashing behavior of fireflies with three newlineimprovements like kernel, updated step size, and random number generation. newlineDynamically create and evaluate instantiations of DevQLMLOps with newlinemethods like Support Vector Machine (SVM), Gradient Boosting Algorithm newline(GBA), and Stochastic Gradient Descent (SGD). AdaBoost has been used to newlinemerge the process of SVM, GBA, and SGD. The results of these classifiers newlineare combined via a voting system. newline