N-Beats Optimization With K-Fold Cross-Validation For Stock Market Price Prediction In Indonesia

Abstract

In the era of capital market digitalization, stock price prediction poses a significant challenge, particularly in developing countries like Indonesia, where high market volatility is driven by political dynamics and exchange rate fluctuations. This study aims to address these challenges by developing a stock price prediction model using the N-BEATS architecture, which is designed to overcome the limitations of traditional methods such as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). The dataset includes open, high, low, close, and volume from five leading Indonesian banks between 2004 and 2024. The N-BEATS model is optimized using K-Fold cross-validation to enhance accuracy and reduce overfitting risks. Evaluation results demonstrate that the N-BEATS model gave better accuracy compared to CNN and RNN models, with a 30% improvement over CNN and 25% over RNN. Analysis of performance variability across stock symbols reveals that the intrinsic data characteristics of each stock influence prediction accuracy. The N-BEATS model exhibits significant potential for stock price prediction in the Indonesian market, excelling in capturing long-term dependencies and offering better interpretability.