An advanced weighted graph algorithm for feature selection in big data

Abstract

With the recent proliferation of intelligent information systems, large data collections containing substantial amounts of repeated and unintentional interference-oriented data are being gathered, resulting in the operation of a vast feature set. However, higher dimensional inputs often include more correlated variables, which can adversely affect model performance. To address this issue and enhance performance, a hybrid feature selection method called Hybrid Binary Gravitational Search Particle Swarm Optimization (HBGSPSO) is developed, which combines the advantages of Binary Gravitational Search Particle Swarm Optimization with an Enhanced Convolution Neural Network Bidirectional Long Short-Term Memory (ECNN-BiLSTM). The Bidirectional Long Short-Term Memory (BiLSTM) is used to extract hidden dynamic data and leverage memory cells to consider long-term historical data following the convolution process. To evaluate the efficiency of the proposed system, numerous experiments are conducted on thirteen well-defined datasets from the machine learning database of UC Irvine. The experiments utilize K Nearest Neighbor (KNN) and Decision Tree (DT) as classifiers to assess the performance of the selected features. The outcomes are compared and contrasted with bio-inspired algorithms Genetic Algorithm (GA), Grey Wolf Optimizer (GWO), and Optimization protocol using Particle Swarm Optimization (PSO). newline