Comparative Performance Analysis of Optimization Algorithms in Artificial Neural Networks for Stock Price Prediction

Abstract

This study aims to enhance price prediction accuracy using Artificial Neural Networks (ANN) by comparing three optimization methods: Stochastic Gradient Descent (SGD), Adam, and RMSprop. The research employs a systematic approach involving the design, training, and validation of ANN models optimized by these techniques. Performance metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE), and R Square are utilized to evaluate the effectiveness of each method. The results indicate that the Adam optimization method outperforms the others, achieving the lowest MSE of 0.0000503 and the lowest MAE of 0.0046, resulting in an impressive R Square value of 0.9989. Adam’s superior performance can be attributed to its adaptive learning rate mechanism, which effectively adjusts to the high volatility and noise characteristic of stock price data, enabling the model to converge faster and more accurately. In comparison, SGD produced a higher MSE of 0.0001208 and MAE of 0.0075, while RMSprop yielded an MSE of 0.0000726 and MAE of 0.0059. These findings highlight Adam’s ability to significantly enhance the predictive capabilities of ANN, particularly in dynamic and complex datasets, making it a preferred choice for this application. The novelty of this research lies not only in its comparative analysis of various optimization methods within the ANN framework but also in the exploration of unique ANN features and their application to a specific stock price prediction case study, providing deeper insights into the practical implications of optimization strategies. This study lays the groundwork for future research by suggesting the exploration of additional optimization algorithms and more complex neural network architectures to further improve prediction accuracy.