Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Boosting the Accuracy of Stock Market Prediction via Multi-Layer Hybrid MTL Structure

Yuxi Hong

TL;DR

This work addresses the challenge of predicting stock market movements in the presence of nonlinear, high-dimensional, and volatile data. It introduces a multi-layer hybrid multi-task learning framework that fuses a Transformer encoder, a Kolmogorov-Arnold Network (KAN), and a BiGRU to jointly forecast multiple targets from multivariate inputs such as Open, High, Low, Close, Volume, and Amount. The approach leverages the Transformer to capture complex feature relationships, KAN for adaptive nonlinear approximation, and BiGRU for long-term temporal dynamics, enabling robust, multi-task predictions with favorable metrics across MAE, MAPE, and $R^2$ compared to strong baselines. Experimental results, including ablation studies and cross-validation, demonstrate superior accuracy and stability, suggesting practical value for informed trading and capital allocation decisions in dynamic markets.

Abstract

Accurate stock market prediction provides great opportunities for informed decision-making, yet existing methods struggle with financial data's non-linear, high-dimensional, and volatile characteristics. Advanced predictive models are needed to effectively address these complexities. This paper proposes a novel multi-layer hybrid multi-task learning (MTL) framework aimed at achieving more efficient stock market predictions. It involves a Transformer encoder to extract complex correspondences between various input features, a Bidirectional Gated Recurrent Unit (BiGRU) to capture long-term temporal relationships, and a Kolmogorov-Arnold Network (KAN) to enhance the learning process. Experimental evaluations indicate that the proposed learning structure achieves great performance, with an MAE as low as 1.078, a MAPE as low as 0.012, and an R^2 as high as 0.98, when compared with other competitive networks.

Boosting the Accuracy of Stock Market Prediction via Multi-Layer Hybrid MTL Structure

TL;DR

This work addresses the challenge of predicting stock market movements in the presence of nonlinear, high-dimensional, and volatile data. It introduces a multi-layer hybrid multi-task learning framework that fuses a Transformer encoder, a Kolmogorov-Arnold Network (KAN), and a BiGRU to jointly forecast multiple targets from multivariate inputs such as Open, High, Low, Close, Volume, and Amount. The approach leverages the Transformer to capture complex feature relationships, KAN for adaptive nonlinear approximation, and BiGRU for long-term temporal dynamics, enabling robust, multi-task predictions with favorable metrics across MAE, MAPE, and compared to strong baselines. Experimental results, including ablation studies and cross-validation, demonstrate superior accuracy and stability, suggesting practical value for informed trading and capital allocation decisions in dynamic markets.

Abstract

Accurate stock market prediction provides great opportunities for informed decision-making, yet existing methods struggle with financial data's non-linear, high-dimensional, and volatile characteristics. Advanced predictive models are needed to effectively address these complexities. This paper proposes a novel multi-layer hybrid multi-task learning (MTL) framework aimed at achieving more efficient stock market predictions. It involves a Transformer encoder to extract complex correspondences between various input features, a Bidirectional Gated Recurrent Unit (BiGRU) to capture long-term temporal relationships, and a Kolmogorov-Arnold Network (KAN) to enhance the learning process. Experimental evaluations indicate that the proposed learning structure achieves great performance, with an MAE as low as 1.078, a MAPE as low as 0.012, and an R^2 as high as 0.98, when compared with other competitive networks.
Paper Structure (17 sections, 14 equations, 13 figures, 4 tables)

This paper contains 17 sections, 14 equations, 13 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Traditional Methods
  4. Machine Learning Networks
  5. Hybrid Models
  6. The Proposed Method
  7. Problem Statement
  8. Overview of the Multi-Layer Hybrid MTL Structure
  9. Transformer Encoder Layer
  10. Detailed KAN Layer
  11. Detailed BiGRU Layer
  12. EXPERIMENTAL EVALUATIONS
  13. Experimental Design
  14. Evaluating Indicator
  15. Result Evaluation and Analysis with Advanced Methods
  16. ...and 2 more sections

Figures (13)

  • Figure 1: The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiGRU.
  • Figure 2: The multi-head attention in the Transformer structure.
  • Figure 3: The detailed comparison of MLP and KAN structure
  • Figure 4: The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiGRU.
  • Figure 5: The multi-layer hybrid MTL structure for stock market prediction, involving an enhanced Transformer, KAN and BiGRU.
  • ...and 8 more figures