Parallel Compact Marine Predators Algorithm-Optimized Backpropagation Neural Network for Enhanced Stock Price Prediction

This paper proposed a novel Parallel Compact Marine Predators Algorithm (PC-MPA) to optimize the weights and biases of a Backpropagation (BP) neural network for stock price prediction, addressing the limitations of traditional gradient-based optimization methods which often converge to suboptimal solutions. The PC-MPA draws inspiration from the foraging behavior of marine predators, compactifying the population into a probability distribution to enhance search efficiency while employing parallelization to accelerate convergence through independent sub-population evolution with periodic information exchange. The BP neural network, structured with input, hidden, and output layers, processes historical stock data to predict future prices, with its performance critically dependent on the optimized parameters derived from PC-MPA. The fitness function, defined as the mean squared error between predicted and actual prices, guides the predator movement in PC-MPA, ensuring iterative refinement of solutions. Furthermore, the integration of PC-MPA with BP neural networks demonstrates superior prediction accuracy compared to conventional approaches, as evidenced by experimental results. The proposed method not only mitigates the risk of local optima but also scales effectively for high-dimensional financial datasets. This work contributes a robust hybrid framework for stock price forecasting, combining metaheuristic optimization with neural networks to improve reliability and computational efficiency. The significance of this approach lies in its potential to support informed decision-making in volatile financial markets, offering a practical tool for investors and analysts.