Optimization of Extreme Learning Machine Using Particle Swarm Algorithm

Employing Particle Swarm Optimization (PSO) to fine-tune Extreme Learning Machine (ELM) parameters significantly enhances prediction accuracy. PSO is a population-based optimization algorithm inspired by collective behaviors like bird flocking or fish schooling. In code implementations, each particle typically represents a candidate solution (e.g., ELM's input weights and biases) moving through a multidimensional search space. The algorithm evaluates particles using fitness functions (e.g., mean squared error) and updates velocities through social and cognitive components using equations like: v_i(t+1) = w*v_i(t) + c1*r1*(pbest_i - x_i(t)) + c2*r2*(gbest - x_i(t)).

By iteratively adjusting ELM's parameters (input weights and hidden layer biases) via PSO, we achieve optimal configuration that boosts model performance. ELM's architectural advantage lies in its single-hidden-layer feedforward network with randomly initialized hidden nodes and analytically calculated output weights. Key implementation steps include: 1) Encoding ELM parameters into particles, 2) Defining fitness evaluation using cross-validation accuracy, 3) Implementing velocity/position update rules. This hybrid approach demonstrates particular efficacy in image processing tasks (e.g., feature selection optimization), pattern recognition systems (parameter-sensitive classifiers), and data mining applications requiring rapid model deployment with high precision.