A feature classification learning method based on multi-objective swarm intelligence optimization

In the context of the exponential growth in data volume, the efficient and accurate classification of large and often redundant datasets has become a critical aspect of data processing and analysis. The Extreme Learning Machine (ELM), known for its randomized parameter selection and efficient hidden layer mapping, is particularly adept at classifying nonlinear problems, especially when dealing with large-scale data. However, the random initialization of hidden layer parameters in a single ELM network introduces inherent uncertainty, which can lead to significant fluctuations in data representation and processing within the network. These fluctuations may result in instability in classification outcomes. Additionally, when the training data contain noise, individual ELMs may be prone to overfitting.To address these challenges, this paper presents a novel enhanced ELM classification algorithm, referred to as ELM-MOFGA. The algorithm utilizes a Multi-Objective Fitness Genetic Algorithm (MOFGA) to evaluate and optimize a series of ELM classifiers, incorporating a more robust double error rate evaluation to improve the stability and reliability of the classifiers. The most optimal predictions from the ensemble of multiple ELM models are then aggregated through a voting mechanism, significantly enhancing the models' overall performance and stability. Experimental results demonstrate that the proposed ELM-MOFGA algorithm achieves superior prediction accuracy when applied to the Cancer, Diabetic, and Fourclass datasets.