Wavelet-CNN for temporal data: Enhancing long-term stock price prediction via multi-resolution wavelet decomposition and CNN-based feature extraction

• This study acknowledges the importance of stock price prediction and its challenges, and addresses them by proposing a new model for stock price prediction. • The proposed model combines multi-resolution analysis with a Convolutional Neural Network (CNN) to efficiently and automatically extract the features of stock prices at different temporal resolutions. • First, we obtain components at different temporal resolutions using wavelet transform. • Subsequently, the wavelet-transformed data is image-transformed, and CNN processing is performed to automatically extract useful temporal features for prediction. • As a result, it is shown that the model we constructed has higher long-term stock price prediction accuracy compared to existing prediction methods. The global economy relies heavily on stock markets, making accurate stock price predictions essential for academic research and practical applications. The task of predicting stock prices presents significant challenges due to the non-linear relationships between historical and future values and the multitude of factors influencing price fluctuations. To address these challenges, we propose an approach that combines wavelet transformation and a convolutional neural network (CNN), both of which are specialized for long-term stock price prediction, to efficiently and automatically extract the features of stock prices at various temporal resolutions. Specifically, we first acquire components with different temporal resolutions using wavelet transform, then convert the wavelet-transformed data into images, and finally perform CNN processing to automatically extract useful temporal features for prediction. Experimental results demonstrate that our method achieves a higher prediction accuracy than conventional machine learning methods, especially in long-term predictions.