GN-CNN-LSTM: Financial Market Prediction With Gaussian Noise Embedded CNN LSTM

Accurate financial market prediction is challenging due to volatility and complex patterns. This study presents a novel approach that injects Gaussian noise into a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) model, enhancing prediction accuracy. While denoising techniques such as Autoencoders, Wavelet Transform, Filtering Techniques, and Generative Adversarial Networks (GANs) are commonly used, this research demonstrates that injecting noise during preprocessing significantly improves the performance of the CNNLSTM model. Empirical results using historical data show that the CNN-LSTM model with injected Gaussian noise outperforms traditional models, with notable improvements in MAE and RMSE metrics. Specifically, in SHOMP, SHI, HSI, and XAUUSD, the value of MAE in different sequences improved to $0.017,0.052,0.030$, and 0.021, respectively, highlighting the effectiveness of the GN-CNN-LSTM model. These findings underscore the potential of noise-injected deep learning models, instilling optimism for achieving more reliable financial market predictions.