Financial Time Series Prediction Using Pelican Optimized Extreme Learning Machine with Reduced Weights

This paper presents a new forecasting approach using the Pelican Optimized Extreme Learning Machine (PO-ELM) model, designed to enhance the prediction of future trends based on historical data. The PO-ELM model refines the ELM by introducing the pelican optimizer, which systematically identifies the decision parameters of the ELM like input weights and biases. In contrast to the conventional ELM, where weights are generated randomly, the PO-ELM method ensures that these parameters are optimized, leading to accurate and reliable forecasts. This enhancement significantly improves the model's predictive capabilities, especially for complex time series data like stock market information. The model is tested on real-time data, and improvements are observed in prediction accuracy via performance metrics such as root mean square error and coefficient of correlation. Additionally, the PO-ELM model achieves these improvements with a reduced number of weights and hidden layer neurons, demonstrating its efficiency in managing computational resources while maintaining high performance. The results underscore the potential of PO-ELM in delivering superior forecasting outcomes and comparisons are performed with ELM and radial basis function neural networks to show the advantages over other methods.