Hybrid lion and exponential PSO-based metaheuristic clustering approach for efficient dynamic data stream management

Area of Science:

• Data Science
• Artificial Intelligence
• Machine Learning

Background:

• Dynamic data streams present challenges for big data exploration due to memory constraints of tracking individual historical data.
• Real-time analysis requires efficient methods for exploring and storing information from historical data in a single pass.
• Existing dynamic clustering algorithms must address concept drift and concept evolution, including changes in attribute associations within clusters.

Purpose of the Study:

• To propose a novel Hybrid Lion and Exponential PSO-based Metaheuristic Clustering Approach (HLEPSOMCA) for efficient dynamic data stream management.
• To ensure the proposed approach satisfies the requirements of concept drift and concept evolution.
• To develop a scalable clustering method with minimized parameters for high-dimensional data.

Main Methods:

• The HLEPSOMCA approach integrates stochastic optimization and deterministic clustering techniques for optimal cluster centering.
• Density clustering strategies are employed to determine micro-clusters.
• Lion and Exponential Particle Swarm Optimization (PSO) are utilized in the initialization phase to maximize performance.

Main Results:

• The HLEPSOMCA approach demonstrated good scalability and required minimal parameters for high-dimensional datasets.
• Experimental results on the KDD-99 dataset showed significant improvements in clustering purity.
• The proposed HLEPSOMCA scheme achieved an average purity improvement of 13.24% compared to baseline approaches.

Conclusions:

• The HLEPSOMCA approach offers an effective solution for dynamic data stream clustering, addressing key challenges like concept drift and evolution.
• The method provides enhanced big data exploration capabilities in real-time scenarios.
• The improved purity validates the efficacy of the hybrid metaheuristic approach in data stream mining.