کنفرانس مهندسی برق ایران

صفحه اصلی / سی و سومین کنفرانس بین المللی مهندسی برق

Integration of Deep Learning Techniques in Stock Market Forecasting: xLSTM-CNN with RevIN and Adaptive Wavelet Denoising

نویسندگان :

Alireza Mohammadi¹ Ali Doustmohammadi² Masoud Shafiee³

1- دانشگاه صنعتی امیرکبیر 2- دانشگاه صنعتی امیرکبیر 3- دانشگاه صنعتی امیرکبیر

کلمات کلیدی :

Time Series،Prediction،Stock Market،xLSTM،CNN،Wavelet Denoising،S&P 500

چکیده :

Stock market prediction is a challenging task due to the nonlinear nature of financial time series data. Recent advances in machine learning and deep learning have offered new solutions to address some of these challenges. This research presents a novel integrated deep learning approach, combining the Extended Long Short-Term Memory (xLSTM) network with a Convolutional Neural Network (CNN) for stock market forecasting. To address the challenge of distribution shift in price signals, the Reversible Instance Normalization (RevIN) method is utilized for efficient input normalization. Additionally, wavelet denoising, combined with an adaptive thresholding technique, is used to mitigate the impact of noise in financial time series. The model is evaluated on the S\&P 500 index, and comparative experiments against advanced forecasting models, such as Extended Long Short-Term Memory for Time Series (xLSTM-TS), Temporal Convolutional Networks (TCN), and transformer-based time series model, demonstrate the superiority of our approach in accurately predicting stock prices and directional movements.